├── shimacos ├── .gitkeep ├── bin │ ├── .gitkeep │ ├── stacking_exp061_030_truncate_cat.sh │ ├── stacking_exp059_030_truncate_lgbm.sh │ └── stacking_exp060_030_truncate_small.sh ├── input ├── components │ ├── __init__.py │ ├── stacking │ │ ├── sync_batchnorm │ │ │ ├── __init__.py │ │ │ ├── unittest.py │ │ │ ├── batchnorm_reimpl.py │ │ │ ├── replicate.py │ │ │ └── comm.py │ │ ├── io.py │ │ └── main_nn.py │ ├── factories │ │ ├── detector_factory.py │ │ ├── __init__.py │ │ ├── collate_factory.py │ │ ├── sampler_factory.py │ │ ├── callback_factory.py │ │ ├── scheduler_factory.py │ │ └── optimizer_factory.py │ └── preprocess │ │ └── make_fold_v2.py ├── yamls │ ├── feature │ │ ├── stacking_030_truncate_small.yaml │ │ └── stacking_009.yaml │ └── stacking.yaml └── .gitignore ├── kami ├── run │ ├── conf │ │ ├── cat_features │ │ │ ├── base.yaml │ │ │ ├── 002.yaml │ │ │ ├── 001.yaml │ │ │ ├── 003.yaml │ │ │ ├── 006.yaml │ │ │ ├── 004.yaml │ │ │ ├── 005.yaml │ │ │ └── 007.yaml │ │ ├── model │ │ │ ├── Spec1D.yaml │ │ │ ├── Spec2DCNN.yaml │ │ │ ├── Spec2DCNNAffine.yaml │ │ │ ├── Spec2DCNNOverlap.yaml │ │ │ ├── Spec2DCNN2Day.yaml │ │ │ ├── Spec2DCNN2DayV2.yaml │ │ │ ├── Spec2DCNNMinMax.yaml │ │ │ ├── Spec2DCNNSplit.yaml │ │ │ ├── Spec2DCNNSplitCat.yaml │ │ │ ├── SpecWeightAvg.yaml │ │ │ └── CenterNet.yaml │ │ ├── decoder │ │ │ ├── MLPDecoder.yaml │ │ │ ├── LSTMDecoder.yaml │ │ │ ├── TransformerDecoder.yaml │ │ │ └── UNet1DDecoder.yaml │ │ ├── loss │ │ │ ├── bce.yaml │ │ │ ├── focal.yaml │ │ │ ├── bi_tempered.yaml │ │ │ ├── tolerance.yaml │ │ │ ├── tolerance_mse.yaml │ │ │ ├── tolerance_nonzero.yaml │ │ │ └── focal_bce.yaml │ │ ├── feature_extractor │ │ │ ├── LSTMFeatureExtractor.yaml │ │ │ ├── SpecFeatureExtractor.yaml │ │ │ ├── CNNSpectrogram.yaml │ │ │ └── PANNsFeatureExtractor.yaml │ │ ├── features │ │ │ ├── 001.yaml │ │ │ ├── base.yaml │ │ │ ├── 002.yaml │ │ │ ├── 003.yaml │ │ │ ├── 005.yaml │ │ │ ├── 004.yaml │ │ │ ├── 006.yaml │ │ │ ├── 007.yaml │ │ │ ├── 009.yaml │ │ │ ├── 010.yaml │ │ │ ├── 011.yaml │ │ │ ├── 008.yaml │ │ │ ├── 012.yaml │ │ │ ├── 013.yaml │ │ │ ├── 014.yaml │ │ │ ├── 015.yaml │ │ │ ├── 016.yaml │ │ │ ├── 017.yaml │ │ │ ├── 021.yaml │ │ │ ├── 025.yaml │ │ │ ├── 018.yaml │ │ │ ├── 024.yaml │ │ │ ├── 031.yaml │ │ │ ├── 022.yaml │ │ │ ├── 023.yaml │ │ │ ├── 027.yaml │ │ │ ├── 028.yaml │ │ │ ├── 029.yaml │ │ │ ├── 030.yaml │ │ │ ├── 032.yaml │ │ │ ├── 019.yaml │ │ │ ├── 020.yaml │ │ │ └── 026.yaml │ │ ├── post_process │ │ │ ├── train.yaml │ │ │ └── score.yaml │ │ ├── ignore │ │ │ ├── 001.yaml │ │ │ ├── 002.yaml │ │ │ └── zero.yaml │ │ ├── dir │ │ │ └── local.yaml │ │ ├── prepare_data.yaml │ │ ├── split_folds.yaml │ │ ├── cv_inference.yaml │ │ ├── cv_train.yaml │ │ └── split │ │ │ ├── stratify_fold_0.yaml │ │ │ └── stratify_fold_1.yaml │ └── split_folds.py ├── src │ ├── models │ │ ├── loss │ │ │ ├── bce.py │ │ │ ├── focal.py │ │ │ ├── focal_bce.py │ │ │ ├── tolerance.py │ │ │ ├── tolerance_mse.py │ │ │ └── tolerance_nonzero.py │ │ ├── decoder │ │ │ ├── mlpdecoder.py │ │ │ ├── lstmdecoder.py │ │ │ └── transformerdecoder.py │ │ ├── feature_extractor │ │ │ ├── spectrogram.py │ │ │ ├── lstm.py │ │ │ ├── panns.py │ │ │ └── cnn.py │ │ ├── spec1D.py │ │ ├── spec2Dcnn.py │ │ ├── spec2DcnnOverlap.py │ │ ├── spec2Dcnn2DayV2.py │ │ ├── spec2DcnnAffine.py │ │ ├── spec2Dcnn2Day.py │ │ └── spec2DcnnMinMax.py │ ├── augmentation │ │ ├── mixup.py │ │ └── cutmix.py │ └── utils │ │ ├── common.py │ │ └── score.py └── .gitignore ├── .gitignore ├── sakami ├── config │ └── meta.yaml └── src │ ├── __init__.py │ ├── config.py │ └── utils.py ├── compose.yaml ├── README.md ├── Dockerfile ├── pyproject.toml ├── requirements.lock ├── requirements-dev.lock └── input └── folds ├── stratify_fold_0.yaml ├── stratify_fold_1.yaml ├── stratify_fold_2.yaml ├── stratify_fold_3.yaml └── stratify_fold_4.yaml /shimacos/.gitkeep: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /shimacos/bin/.gitkeep: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /shimacos/input: -------------------------------------------------------------------------------- 1 | ../input -------------------------------------------------------------------------------- /shimacos/components/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /kami/run/conf/cat_features/base.yaml: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /kami/run/conf/model/Spec1D.yaml: -------------------------------------------------------------------------------- 1 | name: Spec1D -------------------------------------------------------------------------------- /kami/run/conf/decoder/MLPDecoder.yaml: -------------------------------------------------------------------------------- 1 | name: MLPDecoder -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | input 2 | !input/fold/* 3 | output 4 | __pycache__ 5 | -------------------------------------------------------------------------------- /sakami/config/meta.yaml: -------------------------------------------------------------------------------- 1 | # competition 2 | competition_name: . 3 | -------------------------------------------------------------------------------- /kami/run/conf/cat_features/002.yaml: -------------------------------------------------------------------------------- 1 | hour: 2 | dim: 10 3 | num_category: 24 4 | -------------------------------------------------------------------------------- /kami/run/conf/loss/bce.yaml: -------------------------------------------------------------------------------- 1 | name: bce 2 | bce_weight: 1.0 3 | tolerance_weight: 1.0 -------------------------------------------------------------------------------- /kami/run/conf/loss/focal.yaml: -------------------------------------------------------------------------------- 1 | name: focal 2 | 3 | 4 | alpha: 0.5 5 | gamma: 2.0 -------------------------------------------------------------------------------- /kami/run/conf/cat_features/001.yaml: -------------------------------------------------------------------------------- 1 | periodicity: 2 | dim: 4 3 | num_category: 2 4 | -------------------------------------------------------------------------------- /kami/run/conf/cat_features/003.yaml: -------------------------------------------------------------------------------- 1 | minute_15: 2 | dim: 4 3 | num_category: 15 4 | -------------------------------------------------------------------------------- /kami/run/conf/loss/bi_tempered.yaml: -------------------------------------------------------------------------------- 1 | name: bi_tempered 2 | t1: 1.0 3 | t2: 1.0 4 | label_smoothing: 0.0 -------------------------------------------------------------------------------- /kami/run/conf/loss/tolerance.yaml: -------------------------------------------------------------------------------- 1 | name: tolerance 2 | loss_weight: 3 | - 0.5 # bce 4 | - 0.5 # tolerance -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNN.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNN 2 | encoder_name: resnet34 3 | encoder_weights: imagenet -------------------------------------------------------------------------------- /sakami/src/__init__.py: -------------------------------------------------------------------------------- 1 | from .config import Config 2 | 3 | meta_config = Config.load("config/meta.yaml") 4 | -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNNAffine.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNN 2 | encoder_name: resnet34 3 | encoder_weights: imagenet -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNNOverlap.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNN 2 | encoder_name: resnet34 3 | encoder_weights: imagenet -------------------------------------------------------------------------------- /kami/run/conf/loss/tolerance_mse.yaml: -------------------------------------------------------------------------------- 1 | name: tolerance_mse 2 | loss_weight: 3 | - 0.5 # bce 4 | - 0.5 # tolerance -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNN2Day.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNN2Day 2 | encoder_name: resnet34 3 | encoder_weights: imagenet -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNN2DayV2.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNN2DayV2 2 | encoder_name: resnet34 3 | encoder_weights: imagenet -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNNMinMax.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNNMinMax 2 | encoder_name: resnet34 3 | encoder_weights: imagenet -------------------------------------------------------------------------------- /kami/run/conf/loss/tolerance_nonzero.yaml: -------------------------------------------------------------------------------- 1 | name: tolerance_nonzero 2 | loss_weight: 3 | - 0.5 # bce 4 | - 0.5 # tolerance -------------------------------------------------------------------------------- /kami/run/conf/cat_features/006.yaml: -------------------------------------------------------------------------------- 1 | minute_15: 2 | dim: 4 3 | num_category: 15 4 | hour: 5 | dim: 10 6 | num_category: 24 7 | -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNNSplit.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNNSplit 2 | encoder_name: resnet34 3 | encoder_weights: imagenet 4 | n_split: 3 -------------------------------------------------------------------------------- /kami/run/conf/cat_features/004.yaml: -------------------------------------------------------------------------------- 1 | periodicity: 2 | dim: 4 3 | num_category: 2 4 | hour: 5 | dim: 10 6 | num_category: 24 7 | -------------------------------------------------------------------------------- /kami/run/conf/cat_features/005.yaml: -------------------------------------------------------------------------------- 1 | periodicity: 2 | dim: 4 3 | num_category: 2 4 | minute_15: 5 | dim: 4 6 | num_category: 15 7 | -------------------------------------------------------------------------------- /kami/run/conf/decoder/LSTMDecoder.yaml: -------------------------------------------------------------------------------- 1 | name: LSTMDecoder 2 | hidden_size: 128 3 | num_layers: 2 4 | dropout: 0.2 5 | bidirectional: true -------------------------------------------------------------------------------- /kami/run/conf/decoder/TransformerDecoder.yaml: -------------------------------------------------------------------------------- 1 | name: TransformerDecoder 2 | hidden_size: 256 3 | num_layers: 4 4 | nhead: 4 5 | dropout: 0.2 -------------------------------------------------------------------------------- /kami/run/conf/loss/focal_bce.yaml: -------------------------------------------------------------------------------- 1 | name: focal_bce 2 | alpha: 0.5 3 | gamma: 2.0 4 | 5 | weight: 6 | - 1.0 # bce 7 | - 10.0 # focal -------------------------------------------------------------------------------- /kami/run/conf/model/Spec2DCNNSplitCat.yaml: -------------------------------------------------------------------------------- 1 | name: Spec2DCNNSplitCat 2 | encoder_name: resnet34 3 | encoder_weights: imagenet 4 | n_split: 3 -------------------------------------------------------------------------------- /kami/run/conf/model/SpecWeightAvg.yaml: -------------------------------------------------------------------------------- 1 | name: SpecWeightAvg 2 | encoder_name: resnet34 3 | encoder_weights: imagenet 4 | filter_size: 12 -------------------------------------------------------------------------------- /kami/run/conf/decoder/UNet1DDecoder.yaml: -------------------------------------------------------------------------------- 1 | name: UNet1DDecoder 2 | bilinear: false 3 | se: false 4 | res: false 5 | scale_factor: 2 6 | dropout: 0.2 -------------------------------------------------------------------------------- /kami/run/conf/feature_extractor/LSTMFeatureExtractor.yaml: -------------------------------------------------------------------------------- 1 | name: LSTMFeatureExtractor 2 | hidden_size: 64 3 | num_layers: 2 4 | bidirectional: true 5 | -------------------------------------------------------------------------------- /kami/run/conf/feature_extractor/SpecFeatureExtractor.yaml: -------------------------------------------------------------------------------- 1 | name: SpecFeatureExtractor 2 | height: 64 3 | hop_length: ${downsample_rate} 4 | win_length: -------------------------------------------------------------------------------- /kami/run/conf/features/001.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" -------------------------------------------------------------------------------- /kami/run/conf/features/base.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | # - "minute_sin" 8 | # - "minute_cos" -------------------------------------------------------------------------------- /kami/run/conf/cat_features/007.yaml: -------------------------------------------------------------------------------- 1 | minute_15: 2 | dim: 4 3 | num_category: 15 4 | hour: 5 | dim: 10 6 | num_category: 24 7 | periodicity: 8 | dim: 4 9 | num_category: 2 -------------------------------------------------------------------------------- /shimacos/bin/stacking_exp061_030_truncate_cat.sh: -------------------------------------------------------------------------------- 1 | 2 | python -m components.stacking.main_cat \ 3 | feature=stacking_030_truncate \ 4 | store.model_name=$(basename $0 .sh) 5 | 6 | -------------------------------------------------------------------------------- /kami/run/conf/model/CenterNet.yaml: -------------------------------------------------------------------------------- 1 | name: CenterNet 2 | params: 3 | encoder_name: resnet34 4 | encoder_weights: imagenet 5 | keypoint_weight: 1.0 6 | offset_weight: 1.0 7 | bbox_size_weight: 1.0 -------------------------------------------------------------------------------- /kami/run/conf/feature_extractor/CNNSpectrogram.yaml: -------------------------------------------------------------------------------- 1 | name: CNNSpectrogram 2 | base_filters: 64 3 | kernel_sizes: 4 | - 32 5 | - 16 6 | - 2 7 | stride: ${downsample_rate} 8 | sigmoid: true 9 | reinit: true -------------------------------------------------------------------------------- /kami/run/conf/features/002.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | #- "minute_sin" 8 | #- "minute_cos" 9 | - "weekday_sin" 10 | - "weekday_cos" -------------------------------------------------------------------------------- /kami/run/conf/post_process/train.yaml: -------------------------------------------------------------------------------- 1 | score_th: 0.02 2 | distance: 80 3 | remove_periodicity: false 4 | periodicity: 5 | filter_size: 10 6 | #downsample_rate: 15 7 | #split_hour: 8 # 4 にすると周期性を検出できないものがあった 8 | #th: 0.99 -------------------------------------------------------------------------------- /kami/run/conf/post_process/score.yaml: -------------------------------------------------------------------------------- 1 | score_th: 0.0038536 2 | distance: 80 3 | remove_periodicity: false 4 | periodicity: 5 | filter_size: 10 6 | #downsample_rate: 15 7 | #split_hour: 8 # 4 にすると周期性を検出できないものがあった 8 | #th: 0.99 -------------------------------------------------------------------------------- /kami/run/conf/features/003.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | #- "minute_sin" 8 | #- "minute_cos" 9 | # - "weekday_sin" 10 | # - "weekday_cos" 11 | - "lids" -------------------------------------------------------------------------------- /kami/run/conf/features/005.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | #- "minute_sin" 8 | #- "minute_cos" 9 | - "weekday_sin" 10 | - "weekday_cos" 11 | - "periodicity" -------------------------------------------------------------------------------- /kami/run/conf/features/004.yaml: -------------------------------------------------------------------------------- 1 | - "anglez_series_norm" #" anglez" 2 | - "enmo_series_norm" # "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | #- "minute_sin" 8 | #- "minute_cos" 9 | - "weekday_sin" 10 | - "weekday_cos" -------------------------------------------------------------------------------- /kami/run/conf/features/006.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | #- "minute_sin" 8 | #- "minute_cos" 9 | - "minute15_sin" 10 | - "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" -------------------------------------------------------------------------------- /kami/run/conf/features/007.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" -------------------------------------------------------------------------------- /kami/run/conf/feature_extractor/PANNsFeatureExtractor.yaml: -------------------------------------------------------------------------------- 1 | name: PANNsFeatureExtractor 2 | base_filters: 64 3 | kernel_sizes: 4 | - 32 5 | - 16 6 | - ${downsample_rate} 7 | stride: ${downsample_rate} 8 | sigmoid: true 9 | reinit: true 10 | win_length: -------------------------------------------------------------------------------- /kami/run/conf/ignore/001.yaml: -------------------------------------------------------------------------------- 1 | train: 2 | - no 3 | 4 | negative: # ネガティブサンプリングの場合のみスキップ 5 | - 05e1944c3818 6 | - 13b4d6a01d27 7 | - 31011ade7c0a 8 | - a596ad0b82aa 9 | # 2~3割 10 | - 60d31b0bec3b 11 | - 60e51cad2ffb 12 | - 10469f6765bf 13 | - 44a41bba1ee7 -------------------------------------------------------------------------------- /kami/run/conf/features/009.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | - "anglez_diff" 4 | - "enmo_diff" 5 | # - "month_sin" 6 | # - "month_cos" 7 | - "hour_sin" 8 | - "hour_cos" 9 | - "minute_sin" 10 | - "minute_cos" 11 | #- "minute15_sin" 12 | #- "minute15_cos" 13 | - "weekday_sin" 14 | - "weekday_cos" -------------------------------------------------------------------------------- /kami/run/conf/dir/local.yaml: -------------------------------------------------------------------------------- 1 | data_dir: /root/app/input/ 2 | processed_dir: /root/app/kami/processed 3 | output_dir: /root/app/kami/output 4 | model_dir: /root/app/kami/output/train 5 | cv_model_dir: /root/app/kami/output/cv_train 6 | kami_dir: /root/app/kami 7 | input_dir: /root/app/input 8 | sub_dir: ./ 9 | -------------------------------------------------------------------------------- /kami/run/conf/ignore/002.yaml: -------------------------------------------------------------------------------- 1 | train: 2 | # 5割以上欠損 3 | - 05e1944c3818 4 | - 13b4d6a01d27 5 | - 31011ade7c0a 6 | - a596ad0b82aa 7 | # 2~3割 8 | # - 60d31b0bec3b 9 | # - 60e51cad2ffb 10 | # - 10469f6765bf 11 | # - 44a41bba1ee7 12 | 13 | 14 | negative: # ネガティブサンプリングの場合のみスキップ 15 | - "no" -------------------------------------------------------------------------------- /kami/run/conf/features/010.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" -------------------------------------------------------------------------------- /kami/run/conf/ignore/zero.yaml: -------------------------------------------------------------------------------- 1 | train: # 学習データから取り除く 2 | - "zero" 3 | # 5割以上欠損 4 | # - 05e1944c3818 5 | # - 13b4d6a01d27 6 | # - 31011ade7c0a 7 | # - a596ad0b82aa 8 | # 2~3割 9 | # - 60d31b0bec3b 10 | # - 60e51cad2ffb 11 | # - 10469f6765bf 12 | # - 44a41bba1ee7 13 | 14 | negative: # ネガティブサンプリングの場合のみスキップ 15 | - "zero" -------------------------------------------------------------------------------- /shimacos/bin/stacking_exp059_030_truncate_lgbm.sh: -------------------------------------------------------------------------------- 1 | 2 | python -m components.stacking.main_lgbm \ 3 | feature=stacking_030_truncate \ 4 | lgbm.params.objective=binary \ 5 | lgbm.params.metric=binary_logloss \ 6 | store.model_name=$(basename $0 .sh) \ 7 | lgbm.params.scale_pos_weight=1 \ 8 | lgbm.params.is_unbalance=False 9 | 10 | -------------------------------------------------------------------------------- /kami/run/conf/features/011.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" -------------------------------------------------------------------------------- /kami/run/conf/features/008.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "enmo_std_diff12" 14 | - "enmo_std_diff60" 15 | - "enmo_std_diff120" 16 | - "enmo_std_diff360" 17 | - "anglez_std_diff12" 18 | - "anglez_std_diff60" 19 | - "anglez_std_diff120" 20 | - "anglez_std_diff360" -------------------------------------------------------------------------------- /kami/run/conf/features/012.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" -------------------------------------------------------------------------------- /kami/run/conf/prepare_data.yaml: -------------------------------------------------------------------------------- 1 | # ---------- Overriding hydra default configs ---------- 2 | hydra: 3 | job: 4 | name: prepare_data 5 | chdir: true 6 | run: 7 | dir: ${dir.output_dir}/${hydra.job.name}/${hydra.job.override_dirname} 8 | sweep: 9 | dir: ${dir.output_dir}/${hydra.job.name}/ 10 | subdir: ${hydra.job.override_dirname} 11 | 12 | defaults: 13 | - _self_ 14 | - dir: local 15 | 16 | phase: train 17 | 18 | 19 | periodicity: 20 | downsample_rate: 12 21 | stride_min: 3 22 | split_min: 120 -------------------------------------------------------------------------------- /kami/run/conf/split_folds.yaml: -------------------------------------------------------------------------------- 1 | # ---------- Overriding hydra default configs ---------- 2 | hydra: 3 | job: 4 | name: prepare_data 5 | chdir: true 6 | run: 7 | dir: ${dir.output_dir}/${hydra.job.name}/${hydra.job.override_dirname} 8 | sweep: 9 | dir: ${dir.output_dir}/${hydra.job.name}/ 10 | subdir: ${hydra.job.override_dirname} 11 | 12 | defaults: 13 | - _self_ 14 | - dir: local 15 | 16 | phase: train 17 | 18 | 19 | periodicity: 20 | downsample_rate: 12 21 | stride_min: 3 22 | split_min: 120 -------------------------------------------------------------------------------- /kami/src/models/loss/bce.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | 5 | class BCEWithLogitsLoss(nn.Module): 6 | def __init__(self, weight: torch.Tensor = None, pos_weight: torch.Tensor = None): 7 | super(BCEWithLogitsLoss, self).__init__() 8 | self.weight = weight 9 | self.pos_weight = pos_weight 10 | self.loss_fn = nn.BCEWithLogitsLoss(weight=self.weight, pos_weight=self.pos_weight) 11 | 12 | def forward(self, logits, labels, masks): 13 | loss = self.loss_fn(logits, labels) 14 | return loss 15 | -------------------------------------------------------------------------------- /shimacos/components/stacking/sync_batchnorm/__init__.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | # File : __init__.py 3 | # Author : Jiayuan Mao 4 | # Email : maojiayuan@gmail.com 5 | # Date : 27/01/2018 6 | # 7 | # This file is part of Synchronized-BatchNorm-PyTorch. 8 | # https://github.com/vacancy/Synchronized-BatchNorm-PyTorch 9 | # Distributed under MIT License. 10 | 11 | from .batchnorm import SynchronizedBatchNorm1d, SynchronizedBatchNorm2d, SynchronizedBatchNorm3d 12 | from .batchnorm import patch_sync_batchnorm, convert_model 13 | from .replicate import DataParallelWithCallback, patch_replication_callback 14 | -------------------------------------------------------------------------------- /shimacos/components/factories/detector_factory.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torch.nn.functional as F 4 | from omegaconf import OmegaConfig 5 | from mmcv import Config 6 | from mmdet.models import build_detector 7 | 8 | 9 | class Detector(nn.Module): 10 | def __init__(self, config: OmegaConfig) -> None: 11 | detection_config = Config.fromfile(config.detection_config_path) 12 | self.model = build_detector( 13 | detection_config.model, 14 | train_cfg=detection_config.train_cfg, 15 | test_cfg=detection_config.test_cfg, 16 | ) 17 | 18 | -------------------------------------------------------------------------------- /kami/run/conf/features/013.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | - "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "anglez_abs_diff_var_24h" 22 | - "anglez_diff_nonzero_5_var_24h" 23 | - "anglez_diff_nonzero_60_var_24h" 24 | - "lids_var_24h" -------------------------------------------------------------------------------- /shimacos/components/factories/__init__.py: -------------------------------------------------------------------------------- 1 | import warnings 2 | 3 | warnings.simplefilter("ignore") 4 | 5 | from .callback_factory import MyCallback, MyProgressBar 6 | from .collate_factory import text_collate 7 | from .dataset_factory import get_dataset 8 | from .loss_factory import get_loss 9 | from .model_factory import get_model 10 | from .optimizer_factory import get_optimizer 11 | from .sampler_factory import get_sampler 12 | from .scheduler_factory import get_scheduler 13 | 14 | __all__ = ["get_model", "get_dataset", "get_loss", "get_scheduler", "get_optimizer", "get_sampler", "MyProgressBar", "MyCallback", "text_collate"] 15 | -------------------------------------------------------------------------------- /compose.yaml: -------------------------------------------------------------------------------- 1 | version: "3" 2 | services: 3 | kaggle: 4 | build: 5 | context: . 6 | dockerfile: Dockerfile 7 | volumes: 8 | - $PWD/input:/root/app/input 9 | - $PWD/sakami:/root/app/sakami 10 | - $PWD/kami:/root/app/kami 11 | - $PWD/shimacos:/root/app/shimacos 12 | - $PWD/tasks.py:/root/app/tasks.py 13 | working_dir: /root/app 14 | command: bash -c "echo start && /bin/bash" 15 | tty: true 16 | deploy: 17 | resources: 18 | reservations: 19 | devices: 20 | - driver: nvidia 21 | count: 1 22 | capabilities: [gpu] 23 | shm_size: "2gb" 24 | -------------------------------------------------------------------------------- /kami/run/conf/features/014.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/015.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | - "enmo_clip" 27 | #- "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/016.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | #- "periodicity" 20 | #- "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/017.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | #- "anglez_diff_nonzero_5_mean_24h" 17 | #- "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/021.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | #- "anglez_diff_nonzero_5" 14 | #- "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/025.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | #- "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/018.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_log_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_log_12_rolling_std" 24 | - "enmo_log_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/024.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_1" 14 | - "anglez_diff_nonzero_5" 15 | - "anglez_diff_nonzero_60" 16 | - "anglez_abs_diff_mean_24h" 17 | - "anglez_diff_nonzero_5_mean_24h" 18 | - "anglez_diff_nonzero_60_mean_24h" 19 | - "lids_mean_24h" 20 | - "periodicity" 21 | - "non_periodicity" 22 | - "enmo_12_rolling_mean" 23 | - "anglez_12_rolling_std" 24 | - "enmo_12_rolling_std" 25 | - "enmo_12_rolling_max" 26 | - "anglez_diff_5min_median" 27 | #- "enmo_clip" 28 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/031.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_diff_nonzero_5_mean_24h" 16 | - "anglez_diff_nonzero_60_mean_24h" 17 | - "lids_mean_24h" 18 | - "periodicity" 19 | - "non_periodicity" 20 | - "enmo_12_rolling_mean" 21 | - "anglez_12_rolling_std" 22 | - "enmo_12_rolling_std" 23 | - "enmo_12_rolling_max" 24 | - "anglez_diff_5min_median" 25 | #- "enmo_clip" 26 | - "enmo_log" 27 | - "anglez_abs_diff_mean_24h" 28 | - "anglez_abs_diff_60_median" -------------------------------------------------------------------------------- /kami/src/models/loss/focal.py: -------------------------------------------------------------------------------- 1 | from torchvision.ops.focal_loss import sigmoid_focal_loss 2 | import torch 3 | import torch.nn as nn 4 | 5 | # https://pytorch.org/vision/main/_modules/torchvision/ops/focal_loss.html 6 | 7 | 8 | class FocalLoss(nn.Module): 9 | def __init__( 10 | self, 11 | alpha: float, 12 | gamma: float, 13 | ): 14 | super(FocalLoss, self).__init__() 15 | self.alpha = alpha 16 | self.gamma = gamma 17 | 18 | def forward(self, logits, labels, masks): 19 | loss = sigmoid_focal_loss( 20 | logits[:, :, [1, 2]], labels[:, :, [1, 2]], alpha=self.alpha, gamma=self.gamma, reduction="mean" 21 | ) 22 | return loss 23 | -------------------------------------------------------------------------------- /kami/run/conf/features/022.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_diff_nonzero_5_max" 16 | - "anglez_diff_nonzero_60_max" 17 | - "anglez_abs_diff_mean_24h" 18 | - "anglez_diff_nonzero_5_mean_24h" 19 | - "anglez_diff_nonzero_60_mean_24h" 20 | - "lids_mean_24h" 21 | - "periodicity" 22 | - "non_periodicity" 23 | - "enmo_12_rolling_mean" 24 | - "anglez_12_rolling_std" 25 | - "enmo_12_rolling_std" 26 | - "enmo_12_rolling_max" 27 | - "anglez_diff_5min_median" 28 | #- "enmo_clip" 29 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/023.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_diff_nonzero_5_std" 16 | - "anglez_diff_nonzero_60_std" 17 | - "anglez_abs_diff_mean_24h" 18 | - "anglez_diff_nonzero_5_mean_24h" 19 | - "anglez_diff_nonzero_60_mean_24h" 20 | - "lids_mean_24h" 21 | - "periodicity" 22 | - "non_periodicity" 23 | - "enmo_12_rolling_mean" 24 | - "anglez_12_rolling_std" 25 | - "enmo_12_rolling_std" 26 | - "enmo_12_rolling_max" 27 | - "anglez_diff_5min_median" 28 | #- "enmo_clip" 29 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/027.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" 28 | - "pca_0" 29 | - "pca_1" 30 | - "pca_2" 31 | - "pca_3" 32 | - "pca_4" -------------------------------------------------------------------------------- /kami/run/conf/features/028.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_diff_nonzero_5_mean_24h" 16 | - "anglez_diff_nonzero_60_mean_24h" 17 | - "lids_mean_24h" 18 | - "periodicity" 19 | - "non_periodicity" 20 | - "enmo_12_rolling_mean" 21 | - "anglez_12_rolling_std" 22 | - "enmo_12_rolling_std" 23 | - "enmo_12_rolling_max" 24 | - "anglez_diff_5min_median" 25 | #- "enmo_clip" 26 | - "enmo_log" 27 | - "anglez_abs_diff_mean_24h" 28 | - "anglez_abs_diff_1_median" 29 | - "anglez_abs_diff_1_std" 30 | - "anglez_abs_diff_1_max" -------------------------------------------------------------------------------- /kami/run/conf/features/029.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_diff_nonzero_5_mean_24h" 16 | - "anglez_diff_nonzero_60_mean_24h" 17 | - "lids_mean_24h" 18 | - "periodicity" 19 | - "non_periodicity" 20 | - "enmo_12_rolling_mean" 21 | - "anglez_12_rolling_std" 22 | - "enmo_12_rolling_std" 23 | - "enmo_12_rolling_max" 24 | - "anglez_diff_5min_median" 25 | #- "enmo_clip" 26 | - "enmo_log" 27 | - "anglez_abs_diff_mean_24h" 28 | - "anglez_abs_diff_5_median" 29 | - "anglez_abs_diff_5_std" 30 | - "anglez_abs_diff_5_max" -------------------------------------------------------------------------------- /kami/run/conf/features/030.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_diff_nonzero_5_mean_24h" 16 | - "anglez_diff_nonzero_60_mean_24h" 17 | - "lids_mean_24h" 18 | - "periodicity" 19 | - "non_periodicity" 20 | - "enmo_12_rolling_mean" 21 | - "anglez_12_rolling_std" 22 | - "enmo_12_rolling_std" 23 | - "enmo_12_rolling_max" 24 | - "anglez_diff_5min_median" 25 | #- "enmo_clip" 26 | - "enmo_log" 27 | - "anglez_abs_diff_mean_24h" 28 | - "anglez_abs_diff_60_median" 29 | - "anglez_abs_diff_60_std" 30 | - "anglez_abs_diff_60_max" -------------------------------------------------------------------------------- /kami/run/conf/features/032.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_diff_nonzero_5_mean_24h" 16 | - "anglez_diff_nonzero_60_mean_24h" 17 | - "lids_mean_24h" 18 | - "periodicity" 19 | - "non_periodicity" 20 | - "enmo_12_rolling_mean" 21 | - "anglez_12_rolling_std" 22 | - "enmo_12_rolling_std" 23 | - "enmo_12_rolling_max" 24 | - "anglez_diff_5min_median" 25 | #- "enmo_clip" 26 | - "enmo_log" 27 | - "anglez_abs_diff_mean_24h" 28 | #- "anglez_abs_diff_60_median" 29 | - "anglez_abs_diff_60_std" 30 | #- "anglez_abs_diff_60_max" -------------------------------------------------------------------------------- /kami/run/conf/features/019.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "enmo_60_rolling_mean" 26 | - "anglez_60_rolling_std" 27 | - "enmo_60_rolling_std" 28 | - "enmo_60_rolling_max" 29 | - "anglez_diff_5min_median" 30 | #- "enmo_clip" 31 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/020.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "enmo_360_rolling_mean" 26 | - "anglez_360_rolling_std" 27 | - "enmo_360_rolling_std" 28 | - "enmo_360_rolling_max" 29 | - "anglez_diff_5min_median" 30 | #- "enmo_clip" 31 | - "enmo_log" -------------------------------------------------------------------------------- /kami/run/conf/features/026.yaml: -------------------------------------------------------------------------------- 1 | - "anglez" 2 | #- "enmo" 3 | # - "month_sin" 4 | # - "month_cos" 5 | - "hour_sin" 6 | - "hour_cos" 7 | - "minute_sin" 8 | - "minute_cos" 9 | #- "minute15_sin" 10 | #- "minute15_cos" 11 | - "weekday_sin" 12 | - "weekday_cos" 13 | - "anglez_diff_nonzero_5" 14 | - "anglez_diff_nonzero_60" 15 | - "anglez_abs_diff_mean_24h" 16 | - "anglez_diff_nonzero_5_mean_24h" 17 | - "anglez_diff_nonzero_60_mean_24h" 18 | - "lids_mean_24h" 19 | - "periodicity" 20 | - "non_periodicity" 21 | - "enmo_12_rolling_mean" 22 | - "anglez_12_rolling_std" 23 | - "enmo_12_rolling_std" 24 | - "enmo_12_rolling_max" 25 | - "anglez_diff_5min_median" 26 | #- "enmo_clip" 27 | - "enmo_log" 28 | - "pca_0" 29 | - "pca_1" 30 | - "pca_2" 31 | - "pca_3" 32 | - "pca_4" 33 | - "pca_5" 34 | - "pca_6" 35 | - "pca_7" 36 | - "pca_8" 37 | - "pca_9" -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | ## Prepare dataset 2 | 3 | - download from kaggle via kaggle-api. 4 | ```bash 5 | kaggle competitions download -c child-mind-institute-detect-sleep-states -p ./input/ 6 | unzip ./input/child-mind-institute-detect-sleep-states.zip -d ./input 7 | ``` 8 | 9 | ## Training 10 | 11 | - setting up using Docker 12 | 13 | ```sh 14 | docker compose up -d 15 | docker compose exec kaggle /bin/bash 16 | ``` 17 | 18 | - training all 19 | ```sh 20 | inv run-all 21 | ``` 22 | - (Optional) If you want to re-split folds, add `--overwrite-folds` option. 23 | - Not reproducible, so results change every time it is run. 24 | - Default values exists in input/folds which we used 25 | ```sh 26 | inv run-all --overwrite-folds 27 | ``` 28 | 29 | ## Inference 30 | 31 | - Use [this notebook](https://www.kaggle.com/shimacos/kaggle-smi-submission-final) 32 | -------------------------------------------------------------------------------- /kami/src/models/loss/focal_bce.py: -------------------------------------------------------------------------------- 1 | from torchvision.ops.focal_loss import sigmoid_focal_loss 2 | import torch 3 | import torch.nn as nn 4 | 5 | # https://pytorch.org/vision/main/_modules/torchvision/ops/focal_loss.html 6 | 7 | 8 | class FocalBCELoss(nn.Module): 9 | def __init__(self, alpha: float, gamma: float, weight: torch.Tensor = torch.tensor([1.0, 1.0])): 10 | super(FocalBCELoss, self).__init__() 11 | self.alpha = alpha 12 | self.gamma = gamma 13 | self.weight = weight 14 | self.bce_fn = nn.BCEWithLogitsLoss() 15 | 16 | def forward(self, logits, labels, masks): 17 | loss = self.weight[0] * self.bce_fn(logits[:, :, 0], labels[:, :, 0]) + self.weight[1] * sigmoid_focal_loss( 18 | logits[:, :, [1, 2]], labels[:, :, [1, 2]], alpha=self.alpha, gamma=self.gamma, reduction="mean" 19 | ) 20 | return loss 21 | -------------------------------------------------------------------------------- /shimacos/components/stacking/sync_batchnorm/unittest.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | # File : unittest.py 3 | # Author : Jiayuan Mao 4 | # Email : maojiayuan@gmail.com 5 | # Date : 27/01/2018 6 | # 7 | # This file is part of Synchronized-BatchNorm-PyTorch. 8 | # https://github.com/vacancy/Synchronized-BatchNorm-PyTorch 9 | # Distributed under MIT License. 10 | 11 | import unittest 12 | import torch 13 | 14 | 15 | class TorchTestCase(unittest.TestCase): 16 | def assertTensorClose(self, x, y): 17 | adiff = float((x - y).abs().max()) 18 | if (y == 0).all(): 19 | rdiff = 'NaN' 20 | else: 21 | rdiff = float((adiff / y).abs().max()) 22 | 23 | message = ( 24 | 'Tensor close check failed\n' 25 | 'adiff={}\n' 26 | 'rdiff={}\n' 27 | ).format(adiff, rdiff) 28 | self.assertTrue(torch.allclose(x, y), message) 29 | 30 | -------------------------------------------------------------------------------- /shimacos/components/factories/collate_factory.py: -------------------------------------------------------------------------------- 1 | import torch 2 | 3 | 4 | def text_collate(batch): 5 | output = {} 6 | max_length = max([sample["raw_length"] for sample in batch]) 7 | output["id"] = [sample["id"] for sample in batch] 8 | output["input_ids"] = torch.cat( 9 | [torch.as_tensor(sample["input_ids"][:, :max_length]) for sample in batch], 10 | dim=0, 11 | ) 12 | output["attention_mask"] = torch.cat( 13 | [torch.as_tensor(sample["attention_mask"][:, :max_length]) for sample in batch], 14 | dim=0, 15 | ) 16 | output["extension"] = torch.cat( 17 | [torch.as_tensor(sample["extension"]) for sample in batch], 18 | dim=0, 19 | ) 20 | if "label" in batch[0]: 21 | output["label"] = torch.cat( 22 | [torch.as_tensor(sample["label"]) for sample in batch], 23 | dim=0, 24 | ) 25 | return output 26 | -------------------------------------------------------------------------------- /kami/src/models/decoder/mlpdecoder.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torch.nn.functional as F 4 | 5 | 6 | class MLPDecoder(nn.Module): 7 | def __init__(self, n_channels: int, n_classes: int): 8 | super(MLPDecoder, self).__init__() 9 | self.fc1 = nn.Linear(n_channels, 64) 10 | self.fc2 = nn.Linear(64, 64) 11 | self.fc3 = nn.Linear(64, n_classes) 12 | 13 | def forward(self, x: torch.Tensor) -> torch.Tensor: 14 | """Forward pass of the model. 15 | 16 | Args: 17 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 18 | 19 | Returns: 20 | torch.Tensor: (batch_size, n_timesteps, n_classes) 21 | """ 22 | x = x.transpose(1, 2) 23 | x = F.relu(self.fc1(x)) 24 | x = F.dropout(x, p=0.5) 25 | x = F.relu(self.fc2(x)) 26 | x = F.dropout(x, p=0.5) 27 | x = self.fc3(x) 28 | return x 29 | -------------------------------------------------------------------------------- /kami/src/augmentation/mixup.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import torch 3 | 4 | 5 | class Mixup: 6 | def __init__(self, alpha: float = 0.4): 7 | self.alpha = alpha 8 | 9 | def __call__( 10 | self, imgs: torch.Tensor, labels: torch.Tensor 11 | ) -> tuple[torch.Tensor, torch.Tensor]: 12 | """Mixup augmentation. 13 | 14 | Args: 15 | imgs (torch.Tensor): (batch_size, n_channels, n_timesteps) 16 | labels (torch.Tensor): (batch_size, n_timesteps, n_classes) 17 | 18 | Returns: 19 | tuple[torch.Tensor]: mixed_imgs (batch_size, n_channels, n_timesteps) 20 | mixed_labels (batch_size, n_timesteps, n_classes) 21 | """ 22 | batch_size = imgs.size(0) 23 | idx = torch.randperm(batch_size) 24 | lam = np.random.beta(self.alpha, self.alpha) 25 | 26 | mixed_imgs: torch.Tensor = lam * imgs + (1 - lam) * imgs[idx] 27 | mixed_labels: torch.Tensor = lam * labels + (1 - lam) * labels[idx] 28 | 29 | return mixed_imgs, mixed_labels 30 | -------------------------------------------------------------------------------- /shimacos/components/preprocess/make_fold_v2.py: -------------------------------------------------------------------------------- 1 | import os 2 | 3 | import numpy as np 4 | import polars as pl 5 | import yaml 6 | 7 | 8 | def load_data() -> pl.DataFrame: 9 | event_df = pl.read_csv("./input/train_events.csv") 10 | event_df = event_df.with_columns(pl.col("timestamp").str.to_datetime()) 11 | event_df = event_df.with_columns(pl.col("step").cast(pl.UInt32)) 12 | return event_df 13 | 14 | 15 | def main(): 16 | os.makedirs("./input/preprocess", exist_ok=True) 17 | event_df = load_data() 18 | event_df = event_df[["series_id"]].unique() 19 | event_df = event_df.with_columns(pl.Series("row_id", np.arange(len(event_df)))) 20 | 21 | folds = np.zeros(len(event_df)) 22 | for i in range(5): 23 | with open(f"./input/folds/stratify_fold_{i}.yaml", "r") as f: 24 | stratify_fold = yaml.safe_load(f) 25 | valid_ids = stratify_fold["valid_series_ids"] 26 | idx = event_df.filter(pl.col("series_id").is_in(valid_ids))["row_id"].to_numpy() 27 | folds[idx] = i 28 | 29 | event_df = event_df.with_columns(pl.Series("fold", folds)) 30 | event_df[["series_id", "fold"]].write_parquet("./input/preprocess/fold_v2.parquet") 31 | 32 | 33 | if __name__ == "__main__": 34 | main() 35 | -------------------------------------------------------------------------------- /kami/src/models/decoder/lstmdecoder.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | 5 | class LSTMDecoder(nn.Module): 6 | def __init__( 7 | self, 8 | input_size: int, 9 | hidden_size: int, 10 | num_layers: int, 11 | dropout: float, 12 | bidirectional: bool, 13 | n_classes: int, 14 | ): 15 | super().__init__() 16 | self.lstm = nn.LSTM( 17 | input_size=input_size, 18 | hidden_size=hidden_size, 19 | num_layers=num_layers, 20 | dropout=dropout, 21 | bidirectional=bidirectional, 22 | batch_first=True, 23 | ) 24 | hidden_size = hidden_size * 2 if bidirectional else hidden_size 25 | self.linear = nn.Linear(hidden_size, n_classes) 26 | 27 | def forward(self, x: torch.Tensor) -> torch.Tensor: 28 | """Forward pass of the model. 29 | 30 | Args: 31 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 32 | 33 | Returns: 34 | torch.Tensor: (batch_size, n_timesteps, n_classes) 35 | """ 36 | x = x.transpose(1, 2) # (batch_size, n_timesteps, n_channels) 37 | x, _ = self.lstm(x) 38 | x = self.linear(x) 39 | return x 40 | -------------------------------------------------------------------------------- /kami/src/models/loss/tolerance.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | 5 | class ToleranceLoss(nn.Module): 6 | def __init__( 7 | self, 8 | loss_weight: torch.Tensor = torch.Tensor([0.5, 0.5]), 9 | label_weight: torch.Tensor = None, 10 | pos_weight: torch.Tensor = None, 11 | ): 12 | super(ToleranceLoss, self).__init__() 13 | 14 | self.loss_weight = loss_weight 15 | self.label_weight = label_weight 16 | self.pos_weight = pos_weight 17 | 18 | self.loss_fn = nn.BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 19 | 20 | def forward(self, logits, labels, masks): 21 | # logits: shape (batch, seq, class) 22 | # masks: shape (batch, tolerance, seq, class) 23 | 24 | logits_event = logits[:, :, [1, 2]] 25 | 26 | # Step 1: Find the maximum value of logits in the range of each mask 27 | masked_logits_max, _ = torch.max(logits_event.unsqueeze(1) * masks, dim=2) # (batch, tolerance, class) 28 | 29 | # Step 2: Calculate max(0, logits - x) and take the average over seq 30 | loss = torch.relu(logits_event.unsqueeze(1) - masked_logits_max.unsqueeze(2)) # (batch, tolerance, seq, class) 31 | loss = torch.mean(loss) 32 | 33 | loss = self.loss_weight[0] * self.loss_fn(logits, labels) + self.loss_weight[1] * loss 34 | return loss 35 | -------------------------------------------------------------------------------- /kami/src/models/decoder/transformerdecoder.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | 5 | class TransformerDecoder(nn.Module): 6 | def __init__( 7 | self, 8 | input_size: int, 9 | hidden_size: int, 10 | num_layers: int, 11 | dropout: float, 12 | nhead: int, 13 | n_classes: int, 14 | ): 15 | super().__init__() 16 | self.conv = nn.Conv1d(input_size, hidden_size, 1) 17 | transformer_encoder_layer = nn.TransformerEncoderLayer( 18 | d_model=hidden_size, nhead=nhead, dropout=dropout, batch_first=True 19 | ) 20 | self.transformer_encoder = nn.TransformerEncoder( 21 | transformer_encoder_layer, num_layers=num_layers 22 | ) 23 | self.linear = nn.Linear(hidden_size, n_classes) 24 | 25 | def forward(self, x: torch.Tensor) -> torch.Tensor: 26 | """Forward pass of the model. 27 | 28 | Args: 29 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 30 | 31 | Returns: 32 | torch.Tensor: (batch_size, n_timesteps, n_classes) 33 | """ 34 | x = self.conv(x) # (batch_size, n_channels, n_timesteps) 35 | x = x.transpose(1, 2) # (batch_size, n_timesteps, n_channels) 36 | x = self.transformer_encoder(x) 37 | x = self.linear(x) # (batch_size, n_timesteps, n_classes) 38 | 39 | return x 40 | -------------------------------------------------------------------------------- /kami/src/models/loss/tolerance_mse.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | 5 | class ToleranceMSELoss(nn.Module): 6 | def __init__( 7 | self, 8 | loss_weight: torch.Tensor = torch.Tensor([0.5, 0.5]), 9 | label_weight: torch.Tensor = None, 10 | pos_weight: torch.Tensor = None, 11 | ): 12 | super(ToleranceMSELoss, self).__init__() 13 | self.loss_weight = loss_weight 14 | self.label_weight = label_weight 15 | self.pos_weight = pos_weight 16 | 17 | self.loss_fn = nn.BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 18 | 19 | def forward(self, logits, labels, masks): 20 | # logits: shape (batch, seq, class) 21 | # masks: shape (batch, tolerance, seq, class) 22 | 23 | logits_event = logits[:, :, [1, 2]] 24 | 25 | # Step 1: Find the maximum value of logits in the range of each mask 26 | masked_logits_max, _ = torch.max(logits_event.unsqueeze(1) * masks, dim=2) # (batch, tolerance, class) 27 | 28 | # Step 2: Calculate max(0, logits - x) and take the average over seq 29 | loss = torch.square( 30 | torch.relu(logits_event.unsqueeze(1) - masked_logits_max.unsqueeze(2)) 31 | ) # (batch, tolerance, seq, class) 32 | loss = torch.mean(loss) 33 | # ２乗したスケールを元に戻す 34 | loss = torch.sqrt(loss) 35 | 36 | loss = self.loss_weight[0] * self.loss_fn(logits, labels) + self.loss_weight[1] * loss 37 | return loss 38 | -------------------------------------------------------------------------------- /Dockerfile: -------------------------------------------------------------------------------- 1 | FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04 2 | 3 | ENV DEBIAN_FRONTEND=noninteractive 4 | 5 | RUN apt-get update \ 6 | && apt-get install --yes software-properties-common \ 7 | && add-apt-repository ppa:neovim-ppa/stable \ 8 | && apt-get update \ 9 | && apt-get install --yes --no-install-recommends \ 10 | build-essential \ 11 | make \ 12 | curl \ 13 | cmake \ 14 | git \ 15 | libssl-dev \ 16 | libbz2-dev \ 17 | libreadline-dev \ 18 | libsqlite3-dev \ 19 | llvm \ 20 | libncursesw5-dev \ 21 | libxml2-dev \ 22 | libxmlsec1-dev \ 23 | libffi-dev \ 24 | liblzma-dev \ 25 | libgl1-mesa-dev \ 26 | libopencv-dev \ 27 | libtesseract-dev \ 28 | libleptonica-dev \ 29 | tesseract-ocr \ 30 | tesseract-ocr-jpn \ 31 | tesseract-ocr-script-jpan \ 32 | tesseract-ocr-script-jpan-vert \ 33 | tk-dev \ 34 | xz-utils \ 35 | wget \ 36 | zip \ 37 | zlib1g-dev \ 38 | && apt-get clean \ 39 | && rm -rf /var/lib/apt/lists/* 40 | 41 | # python 42 | RUN git clone https://github.com/pyenv/pyenv.git ~/.pyenv \ 43 | && ~/.pyenv/plugins/python-build/bin/python-build 3.10.11 /usr/local/python3.10.11 \ 44 | && rm -r ~/.pyenv 45 | ENV PATH=/usr/local/python3.10.11/bin:$PATH 46 | 47 | # Rye 48 | ENV RYE_HOME="/opt/rye" 49 | ENV PATH="$RYE_HOME/shims:$PATH" 50 | RUN curl -sSf https://rye-up.com/get | RYE_NO_AUTO_INSTALL=1 RYE_INSTALL_OPTION="--yes" bash 51 | 52 | WORKDIR /root/app 53 | COPY README.md pyproject.toml requirements-dev.lock requirements.lock /root/app/ 54 | 55 | RUN rye sync --no-lock 56 | 57 | ENTRYPOINT ["rye" , "run"] 58 | -------------------------------------------------------------------------------- /kami/src/models/feature_extractor/spectrogram.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | 3 | import torch 4 | import torch.nn as nn 5 | import torchaudio.transforms as T 6 | 7 | 8 | class SpecNormalize(nn.Module): 9 | def __init__(self, eps: float = 1e-8): 10 | super().__init__() 11 | self.eps = eps 12 | 13 | def forward(self, x): 14 | # batch, channel毎に正規化 15 | # x: (batch, channel, freq, time) 16 | min_ = x.min(dim=-1, keepdim=True)[0].min(dim=-2, keepdim=True)[0] 17 | max_ = x.max(dim=-1, keepdim=True)[0].max(dim=-2, keepdim=True)[0] 18 | 19 | return (x - min_) / (max_ - min_ + self.eps) 20 | 21 | 22 | class SpecFeatureExtractor(nn.Module): 23 | def __init__( 24 | self, 25 | in_channels: int, 26 | height: int, 27 | hop_length: int, 28 | win_length: Optional[int] = None, 29 | out_size: Optional[int] = None, 30 | ): 31 | super().__init__() 32 | self.height = height 33 | self.out_chans = in_channels 34 | n_fft = height * 2 - 1 35 | self.feature_extractor = nn.Sequential( 36 | T.Spectrogram(n_fft=n_fft, hop_length=hop_length, win_length=win_length), 37 | T.AmplitudeToDB(top_db=80), 38 | SpecNormalize(), 39 | ) 40 | self.out_size = out_size 41 | 42 | if self.out_size is not None: 43 | self.pool = nn.AdaptiveAvgPool2d((None, self.out_size)) 44 | 45 | def forward(self, x: torch.Tensor) -> torch.Tensor: 46 | img = self.feature_extractor(x) 47 | if self.out_size is not None: 48 | img = self.pool(img) 49 | 50 | return img 51 | -------------------------------------------------------------------------------- /kami/run/conf/cv_inference.yaml: -------------------------------------------------------------------------------- 1 | # ---------- Overriding hydra default configs ---------- 2 | hydra: 3 | job: 4 | name: cv_inference 5 | chdir: true 6 | run: 7 | dir: ${dir.output_dir}/${hydra.job.name}/${exp_name}/single 8 | sweep: 9 | dir: ${dir.output_dir}/${hydra.job.name}/${exp_name} 10 | subdir: run${hydra.job.num} 11 | 12 | defaults: 13 | - _self_ 14 | - dir: local 15 | - model: Spec2DCNN 16 | - feature_extractor: CNNSpectrogram 17 | - decoder: UNet1DDecoder 18 | - features: base 19 | - post_process: score 20 | # phase: train 用 21 | - split@fold_0: stratify_fold_0 22 | - split@fold_1: stratify_fold_1 23 | - split@fold_2: stratify_fold_2 24 | - split@fold_3: stratify_fold_3 25 | - split@fold_4: stratify_fold_4 26 | - loss: bce 27 | 28 | datamodule: 29 | how: random # random, stride, overlap 30 | train_stride: 3600 # 3h=2160, 5h=3600, 8h=5760 31 | overlap: 0 32 | zero_periodicity: false 33 | 34 | debug: false 35 | 36 | num_fold: 5 37 | num_tta: 2 38 | 39 | exp_name: dummy 40 | phase: test 41 | weight: 42 | exp_name: ${exp_name} 43 | run_name: cv 44 | 45 | 46 | how_post_process: 'peaks' # group_by_day 47 | 48 | seed: 42 49 | duration: 5760 # durationは32の倍数 50 | downsample_rate: 2 # durationをdownsample_rateで割った値がnum_framesになる 51 | upsample_rate: 1 52 | batch_size: 32 53 | num_workers: 2 54 | use_amp: true 55 | 56 | # augmentation 57 | augmentation: 58 | mixup_prob: 0.0 59 | mixup_alpha: 0.4 60 | cutmix_prob: 0.0 61 | cutmix_alpha: 0.4 62 | 63 | 64 | # label 65 | labels: 66 | - "awake" 67 | - "event_onset" 68 | - "event_wakeup" 69 | label_weight: 70 | - 1.0 71 | - 1.0 72 | - 1.0 73 | pos_weight: 74 | - 1.0 75 | - 1.0 76 | - 1.0 77 | 78 | -------------------------------------------------------------------------------- /kami/src/models/loss/tolerance_nonzero.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | 5 | class ToleranceNonZeroLoss(nn.Module): 6 | def __init__( 7 | self, 8 | loss_weight: torch.Tensor = torch.Tensor([0.5, 0.5]), 9 | label_weight: torch.Tensor = None, 10 | pos_weight: torch.Tensor = None, 11 | ): 12 | super(ToleranceNonZeroLoss, self).__init__() 13 | 14 | self.loss_weight = loss_weight 15 | self.label_weight = label_weight 16 | self.pos_weight = pos_weight 17 | 18 | self.loss_fn = nn.BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 19 | 20 | def forward(self, logits, labels, masks): 21 | # logits: shape (batch, seq, class) 22 | # masks: shape (batch, tolerance, seq, class) 23 | 24 | logits_event = logits[:, :, [1, 2]] 25 | 26 | # Step 1: Find the maximum value of logits in the range of each mask 27 | masked_logits_max, _ = torch.max(logits_event.unsqueeze(1) * masks, dim=2) # (batch, tolerance, class) 28 | 29 | # Step 2: Calculate max(0, logits - x) and take the average over seq 30 | loss = torch.relu(logits_event.unsqueeze(1) - masked_logits_max.unsqueeze(2)) # (batch, tolerance, seq, class) 31 | 32 | # Count the number of positive elements for each sequence 33 | num_positive_elements_seq = torch.sum(loss > 0, dim=2) # (batch, tolerance, class) 34 | # Compute the average loss for each sequence, avoiding division by zero 35 | loss_avg_seq = torch.sum(loss, dim=2) / ( 36 | num_positive_elements_seq + (num_positive_elements_seq == 0) 37 | ) # (batch, tolerance, seq) 38 | loss = torch.mean(loss_avg_seq) 39 | 40 | loss = self.loss_weight[0] * self.loss_fn(logits, labels) + self.loss_weight[1] * loss 41 | return loss 42 | -------------------------------------------------------------------------------- /shimacos/bin/stacking_exp060_030_truncate_small.sh: -------------------------------------------------------------------------------- 1 | 2 | for n_fold in 0 1 2 3 4; do 3 | python -m components.stacking.main_nn \ 4 | base.use_transformer_parameter=False \ 5 | data.n_fold=$n_fold \ 6 | data.is_train=True \ 7 | data.normalize=False \ 8 | data.use_multi_label=True \ 9 | model.model_class=CNNRNNWoDownSample \ 10 | model.num_label=2 \ 11 | base.loss_class=nn.BCEWithLogitsLoss \ 12 | train.warm_start=False \ 13 | store.model_name=$(basename $0 .sh) \ 14 | feature=stacking_030_truncate_small \ 15 | data.dataset_class=StackingDataset \ 16 | train.loss_weight=1 \ 17 | train.batch_size=32 \ 18 | train.epoch=10 \ 19 | train.warmup_multiplier=100 \ 20 | train.learning_rate=0.000001 \ 21 | trainer.precision=16 \ 22 | data.chunk_size=120 \ 23 | data.window_size=60 \ 24 | model.num_layers=1 \ 25 | model.hidden_size=96 \ 26 | model.text.backbone=microsoft/deberta-v3-small \ 27 | model.text.pretrained=False \ 28 | data.use_only_positive_chunk=False \ 29 | data.downsample_feature=False \ 30 | data.is_stacking=True \ 31 | trainer.gradient_clip_val=0.5 \ 32 | base.opt_class=Adam \ 33 | base.distance=8 \ 34 | base.height=0.001 \ 35 | train.refine_target=False \ 36 | train.use_gaussian_target=False \ 37 | trainer.num_sanity_val_steps=5 \ 38 | model.encoder=mean \ 39 | model.decoder=kernel \ 40 | model.kernel_sizes=['3'] 41 | done 42 | -------------------------------------------------------------------------------- /shimacos/yamls/feature/stacking_030_truncate_small.yaml: -------------------------------------------------------------------------------- 1 | cat_cols: 2 | - time_idx 3 | - hour 4 | - minute 5 | - minute_mod15 6 | - weekday 7 | label_col: label 8 | label_cols: 9 | - label_onset 10 | - label_wakeup 11 | numerical_cols: 12 | - pred_onset_148_gru_scale_factor 13 | - pred_onset_156_gru_transformer_residual 14 | - pred_onset_163_gru_sleep_target 15 | - pred_onset_179_gru_minute_embedding_sync 16 | - pred_onset_exp068_transformer 17 | - pred_onset_exp078_lstm 18 | - pred_onset_exp081_mixup_short_feat14 19 | - pred_wakeup_148_gru_scale_factor 20 | - pred_wakeup_156_gru_transformer_residual 21 | - pred_wakeup_163_gru_sleep_target 22 | - pred_wakeup_179_gru_minute_embedding_sync 23 | - pred_wakeup_exp068_transformer 24 | - pred_wakeup_exp078_lstm 25 | - pred_wakeup_exp081_mixup_short_feat14 26 | - pred_sleep_exp068_transformer 27 | - pred_sleep_exp078_lstm 28 | - pred_sleep_exp081_mixup_short_feat14 29 | - pred_sleep_163_gru_sleep_target 30 | - pred_onset_148_gru_scale_factor_raw 31 | - pred_onset_156_gru_transformer_residual_raw 32 | - pred_onset_163_gru_sleep_target_raw 33 | - pred_onset_179_gru_minute_embedding_sync_raw 34 | - pred_onset_exp068_transformer_raw 35 | - pred_onset_exp078_lstm_raw 36 | - pred_onset_exp081_mixup_short_feat14_raw 37 | - pred_wakeup_148_gru_scale_factor_raw 38 | - pred_wakeup_156_gru_transformer_residual_raw 39 | - pred_wakeup_163_gru_sleep_target_raw 40 | - pred_wakeup_179_gru_minute_embedding_sync_raw 41 | - pred_wakeup_exp068_transformer_raw 42 | - pred_wakeup_exp078_lstm_raw 43 | - pred_wakeup_exp081_mixup_short_feat14_raw 44 | - prediction_onset 45 | - prediction_onset_min 46 | - prediction_onset_max 47 | - prediction_wakeup 48 | - prediction_wakeup_min 49 | - prediction_wakeup_max 50 | - prediction_onset_var 51 | - prediction_wakeup_var 52 | pred_col: label_pred 53 | pred_cols: 54 | - label_onset_pred 55 | - label_wakeup_pred 56 | version: 030_truncate 57 | -------------------------------------------------------------------------------- /kami/src/models/feature_extractor/lstm.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | 3 | import torch 4 | import torch.nn as nn 5 | 6 | 7 | class LSTMFeatureExtractor(nn.Module): 8 | def __init__( 9 | self, 10 | in_channels, 11 | hidden_size, 12 | num_layers, 13 | bidirectional, 14 | out_size: Optional[int] = None, 15 | ): 16 | super().__init__() 17 | self.fc = nn.Linear(in_channels, hidden_size) 18 | self.height = hidden_size * (2 if bidirectional else 1) 19 | self.lstm = nn.LSTM( 20 | input_size=hidden_size, 21 | hidden_size=hidden_size, 22 | num_layers=num_layers, 23 | bidirectional=bidirectional, 24 | batch_first=True, 25 | ) 26 | self.out_chans = 1 27 | self.out_size = out_size 28 | if self.out_size is not None: 29 | self.pool = nn.AdaptiveAvgPool2d((None, self.out_size)) 30 | 31 | def forward(self, x: torch.Tensor) -> torch.Tensor: 32 | """Forward pass 33 | 34 | Args: 35 | x (torch.Tensor): (batch_size, in_channels, time_steps) 36 | 37 | Returns: 38 | torch.Tensor: (batch_size, out_chans, height, time_steps) 39 | """ 40 | # x: (batch_size, in_channels, time_steps) 41 | if self.out_size is not None: 42 | x = x.unsqueeze(1) # x: (batch_size, 1, in_channels, time_steps) 43 | x = self.pool(x) # x: (batch_size, 1, in_channels, output_size) 44 | x = x.squeeze(1) # x: (batch_size, in_channels, output_size) 45 | x = x.transpose(1, 2) # x: (batch_size, output_size, in_channels) 46 | x = self.fc(x) # x: (batch_size, output_size, hidden_size) 47 | x, _ = self.lstm(x) # x: (batch_size, output_size, hidden_size * num_directions) 48 | x = x.transpose(1, 2) # x: (batch_size, hidden_size * num_directions, output_size) 49 | x = x.unsqueeze(1) # x: (batch_size, out_chans, hidden_size * num_directions, time_steps) 50 | return x 51 | -------------------------------------------------------------------------------- /shimacos/components/stacking/io.py: -------------------------------------------------------------------------------- 1 | import polars as pl 2 | 3 | 4 | def load_sakami_data(exp_names: list[str]) -> pl.LazyFrame: 5 | dfs = [] 6 | for exp_name in exp_names: 7 | df = pl.read_parquet(f"../sakami/output/{exp_name}/valid_preds.parquet").sort("row_id").drop("row_id") 8 | if "sleep" in df.columns: 9 | df = df.rename( 10 | { 11 | "prediction_wakeup": f"pred_wakeup_{exp_name}", 12 | "prediction_onset": f"pred_onset_{exp_name}", 13 | "sleep": f"pred_sleep_{exp_name}", 14 | } 15 | ) 16 | else: 17 | df = df.rename( 18 | { 19 | "prediction_wakeup": f"pred_wakeup_{exp_name}", 20 | "prediction_onset": f"pred_onset_{exp_name}", 21 | } 22 | ) 23 | dfs.append(df) 24 | df = pl.concat(dfs, how="horizontal") 25 | return df 26 | 27 | 28 | def load_kami_data(exp_names: list[str]) -> pl.LazyFrame: 29 | df = pl.concat( 30 | [ 31 | pl.read_parquet(f"../kami/output/cv_inference/{exp_name}/single/train_pred.parquet").rename( 32 | { 33 | "pred_sleep": f"pred_sleep_{exp_name}", 34 | "pred_wakeup": f"pred_wakeup_{exp_name}", 35 | "pred_onset": f"pred_onset_{exp_name}", 36 | } 37 | ) 38 | for exp_name in exp_names 39 | ], 40 | how="horizontal", 41 | ) 42 | return df 43 | 44 | 45 | def load_shimacos_data(exp_names: list[str]) -> pl.LazyFrame: 46 | df = pl.concat( 47 | [ 48 | pl.concat([pl.read_parquet(f"./output/{exp_name}/fold{i}/result/valid.parquet") for i in range(5)]) 49 | .rename( 50 | { 51 | "label_onset_pred": f"pred_onset_{exp_name}", 52 | "label_wakeup_pred": f"pred_wakeup_{exp_name}", 53 | } 54 | ) 55 | .drop(["label_onset", "label_wakeup"]) 56 | for exp_name in exp_names 57 | ], 58 | how="align", 59 | ) 60 | return df 61 | -------------------------------------------------------------------------------- /pyproject.toml: -------------------------------------------------------------------------------- 1 | [project] 2 | name = "kaggle-cmi" 3 | version = "0.1.0" 4 | description = "Add your description here" 5 | dependencies = [ 6 | "numpy>=1.25.2", 7 | "pandas>=2.0.3", 8 | "polars>=0.19", 9 | "jupyter>=1.0.0", 10 | "matplotlib>=3.7.2", 11 | "seaborn>=0.12.2", 12 | "transformers>=4.31.0", 13 | "pillow==9.5.0", 14 | "torch==2.0.1+cu118", 15 | "torchvision==0.15.2+cu118", 16 | "torchaudio==2.0.2+cu118", 17 | "lightgbm>=4.0.0", 18 | "catboost>=1.2", 19 | "xgboost>=1.7.6", 20 | "hydra-core>=1.3.2", 21 | "hydra_colorlog>=1.2.0", 22 | "google-cloud-storage>=2.10.0", 23 | "deepspeed>=0.10.0", 24 | "seqeval>=1.2.2", 25 | "openmim>=0.3.9", 26 | "fairscale>=0.4.13", 27 | "wandb>=0.15.8", 28 | "pytorch-lightning==2.0.6", 29 | "timm==0.9.2", 30 | "pyarrow>=14.0.1", 31 | "matplotlib-venn>=0.11.9", 32 | "scipy>=1.11.3", 33 | "numba>=0.58.1", 34 | "loguru>=0.7.2", 35 | "segmentation_models_pytorch>=0.3.3", 36 | "invoke>=2.2.0", 37 | "pathlib>=1.0.1", 38 | ] 39 | readme = "README.md" 40 | requires-python = ">= 3.10" 41 | 42 | [build-system] 43 | requires = ["hatchling"] 44 | build-backend = "hatchling.build" 45 | 46 | [tool.hatch.build.targets.wheel] 47 | packages = ["*"] 48 | 49 | [tool.rye] 50 | managed = true 51 | dev-dependencies = [ 52 | "black>=23.7.0", 53 | "mypy>=1.5.0", 54 | "flake8>=6.1.0", 55 | "ruff>=0.0.284", 56 | "kaggle>=1.5.16", 57 | ] 58 | 59 | [tool.hatch.metadata] 60 | allow-direct-references = true 61 | 62 | [[tool.rye.sources]] 63 | name = "torch" 64 | url = "https://download.pytorch.org/whl/cu118" 65 | type = "index" 66 | 67 | [[tool.rye.sources]] 68 | name = "torchvision" 69 | url = "https://download.pytorch.org/whl/cu118" 70 | type = "index" 71 | 72 | [[tool.rye.sources]] 73 | name = "cuml-cu11" 74 | url = "https://pypi.nvidia.com" 75 | type = "index" 76 | 77 | 78 | [tool.ruff] 79 | target-version = "py311" 80 | line-length = 100 81 | ignore = [ 82 | "E402", # Module level import not at top of file 83 | "E501", # Line too long 84 | ] 85 | select = [ 86 | "F", 87 | "E", 88 | "W", 89 | "I", 90 | "B", 91 | ] 92 | -------------------------------------------------------------------------------- /kami/src/augmentation/cutmix.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import torch 3 | 4 | 5 | def get_rand_1dbbox(n_timesteps: int, lam: float) -> tuple[int, int]: 6 | """Get random 1D bounding box. 7 | 8 | Args: 9 | n_timesteps (int): Number of timesteps. 10 | lam (float): Lambda value. 11 | 12 | Returns: 13 | tuple[int, int]: (start, end) of the bounding box. 14 | """ 15 | cut_rat = np.sqrt(1.0 - lam) 16 | cut_len = int(n_timesteps * cut_rat) 17 | 18 | start = np.random.randint(0, n_timesteps - cut_len) 19 | end = start + cut_len 20 | 21 | return start, end 22 | 23 | 24 | class Cutmix: 25 | def __init__(self, alpha: float = 0.4): 26 | self.alpha = alpha 27 | 28 | def __call__( 29 | self, imgs: torch.Tensor, labels: torch.Tensor 30 | ) -> tuple[torch.Tensor, torch.Tensor]: 31 | """Cutmix augmentation. 32 | 33 | Args: 34 | imgs (torch.Tensor): (batch_size, n_channels, n_timesteps) 35 | labels (torch.Tensor): (batch_size, n_timesteps, n_classes) 36 | 37 | Returns: 38 | tuple[torch.Tensor]: mixed_imgs (batch_size, n_channels, n_timesteps) 39 | mixed_labels (batch_size, n_timesteps, n_classes) 40 | """ 41 | batch_size = imgs.size(0) 42 | idx = torch.randperm(batch_size) 43 | 44 | shuffled_imgs = imgs[idx] 45 | shuffled_labels = labels[idx] 46 | 47 | lam = np.random.beta(self.alpha, self.alpha) 48 | start, end = get_rand_1dbbox(imgs.size(2), lam) 49 | 50 | mixed_imgs = torch.concatenate( 51 | [imgs[:, :, :start], shuffled_imgs[:, :, start:end], imgs[:, :, end:]], dim=2 52 | ) 53 | mixed_labels = torch.concatenate( 54 | [labels[:, :start, :], shuffled_labels[:, start:end, :], labels[:, end:, :]], dim=1 55 | ) 56 | 57 | return mixed_imgs, mixed_labels 58 | 59 | 60 | if __name__ == "__main__": 61 | imgs = torch.randn(2, 3, 100) 62 | labels = torch.randn(2, 100, 5) 63 | cutmix = Cutmix() 64 | 65 | mixed_imgs, mixed_labels = cutmix(imgs, labels) 66 | 67 | print(mixed_imgs.shape) 68 | print(mixed_labels.shape) 69 | -------------------------------------------------------------------------------- /kami/src/models/feature_extractor/panns.py: -------------------------------------------------------------------------------- 1 | from typing import Callable, Optional 2 | 3 | import torch 4 | import torch.nn as nn 5 | 6 | from src.models.feature_extractor.cnn import CNNSpectrogram 7 | from src.models.feature_extractor.spectrogram import SpecFeatureExtractor 8 | 9 | 10 | class PANNsFeatureExtractor(nn.Module): 11 | def __init__( 12 | self, 13 | in_channels: int = 3, 14 | base_filters: int | tuple = 128, 15 | kernel_sizes: tuple = (32, 16, 4, 2), 16 | stride: int = 4, 17 | sigmoid: bool = False, 18 | output_size: Optional[int] = None, 19 | conv: Callable = nn.Conv1d, 20 | reinit: bool = True, 21 | win_length: Optional[int] = None, 22 | ): 23 | super().__init__() 24 | self.cnn_feature_extractor = CNNSpectrogram( 25 | in_channels=in_channels, 26 | base_filters=base_filters, 27 | kernel_sizes=kernel_sizes, 28 | stride=stride, 29 | sigmoid=sigmoid, 30 | output_size=output_size, 31 | conv=conv, 32 | reinit=reinit, 33 | ) 34 | self.spec_feature_extractor = SpecFeatureExtractor( 35 | in_channels=in_channels, 36 | height=self.cnn_feature_extractor.height, 37 | hop_length=stride, 38 | win_length=win_length, 39 | out_size=output_size, 40 | ) 41 | self.height = self.cnn_feature_extractor.height 42 | self.out_chans = self.cnn_feature_extractor.out_chans + in_channels 43 | 44 | def forward(self, x: torch.Tensor) -> torch.Tensor: 45 | """Forward pass of the model. 46 | 47 | Args: 48 | x (torch.Tensor): (batch_size, in_channels, time_steps) 49 | 50 | Returns: 51 | torch.Tensor : (batch_size, out_chans, height, time_steps) 52 | """ 53 | 54 | cnn_img = self.cnn_feature_extractor(x) # (batch_size, cnn_chans, height, time_steps) 55 | spec_img = self.spec_feature_extractor(x) # (batch_size, in_channels, height, time_steps) 56 | 57 | img = torch.cat([cnn_img, spec_img], dim=1) # (batch_size, out_chans, height, time_steps) 58 | 59 | return img 60 | -------------------------------------------------------------------------------- /kami/run/conf/cv_train.yaml: -------------------------------------------------------------------------------- 1 | # ---------- Overriding hydra default configs ---------- 2 | hydra: 3 | job: 4 | name: cv_train 5 | chdir: true 6 | run: 7 | dir: ${dir.output_dir}/${hydra.job.name}/${exp_name}/cv 8 | sweep: 9 | dir: ${dir.output_dir}/${hydra.job.name}/${exp_name} 10 | subdir: run${hydra.job.num} 11 | 12 | defaults: 13 | - _self_ 14 | - dir: local 15 | - model: Spec2DCNN # Spec2DCNN2day 16 | - feature_extractor: CNNSpectrogram 17 | - decoder: UNet1DDecoder 18 | - split@fold_0: stratify_fold_0 19 | - split@fold_1: stratify_fold_1 20 | - split@fold_2: stratify_fold_2 21 | - split@fold_3: stratify_fold_3 22 | - split@fold_4: stratify_fold_4 23 | - features: base 24 | - post_process: train 25 | - ignore: zero 26 | - loss: bce 27 | - cat_features: base 28 | 29 | 30 | datamodule: 31 | how: random # random, stride, overlap 32 | train_stride: 3600 # 3h=2160, 5h=3600, 8h=5760 33 | overlap: 0 34 | zero_periodicity: false 35 | max_label_smoothing: 1.0 36 | 37 | label_correct: 38 | use: false 39 | save_epoch: 0 40 | pred_threshold: 0.5 41 | pred_rate: 1.0 42 | 43 | num_fold: 5 44 | 45 | seed: 42 46 | exp_name: dummy 47 | duration: 5760 # (1step 5秒なので) duration * 5 = 秒数。 eg) 5760 * 5 = 28800s = 8h 48 | downsample_rate: 2 49 | upsample_rate: 1 50 | 51 | # training 52 | epoch: 50 53 | batch_size: 32 54 | num_workers: 8 55 | accelerator: auto 56 | use_amp: true 57 | debug: false 58 | gradient_clip_val: 1.0 59 | accumulate_grad_batches: 1 60 | monitor: val_loss 61 | monitor_mode: min 62 | check_val_every_n_epoch: 1 63 | 64 | # Dataset 65 | offset: 10 66 | sigma: 10 67 | bg_sampling_rate: 0.5 68 | sigma_decay: # 0.95 69 | sleep_decay: # 0.90 70 | 71 | # augmentation 72 | augmentation: 73 | mixup_prob: 0.0 74 | mixup_alpha: 0.4 75 | cutmix_prob: 0.0 76 | cutmix_alpha: 0.4 77 | 78 | averaged_model: 79 | how: None # ema, avg 80 | ema_decay: 0.9999 81 | 82 | 83 | # label 84 | labels: 85 | - "awake" 86 | - "event_onset" 87 | - "event_wakeup" 88 | label_weight: 89 | - 1.0 90 | - 1.0 91 | - 1.0 92 | pos_weight: 93 | - 1.0 94 | - 1.0 95 | - 1.0 96 | 97 | 98 | # optimizer 99 | optimizer: 100 | lr: 0.0005 101 | 102 | # scheduler 103 | scheduler: 104 | use_warmup: False 105 | -------------------------------------------------------------------------------- /shimacos/.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | wheels/ 23 | pip-wheel-metadata/ 24 | share/python-wheels/ 25 | *.egg-info/ 26 | .installed.cfg 27 | *.egg 28 | MANIFEST 29 | 30 | # PyInstaller 31 | # Usually these files are written by a python script from a template 32 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 33 | *.manifest 34 | *.spec 35 | 36 | # Installer logs 37 | pip-log.txt 38 | pip-delete-this-directory.txt 39 | 40 | # Unit test / coverage reports 41 | htmlcov/ 42 | .tox/ 43 | .nox/ 44 | .coverage 45 | .coverage.* 46 | .cache 47 | nosetests.xml 48 | coverage.xml 49 | *.cover 50 | *.py,cover 51 | .hypothesis/ 52 | .pytest_cache/ 53 | 54 | # Translations 55 | *.mo 56 | *.pot 57 | 58 | # Django stuff: 59 | *.log 60 | local_settings.py 61 | db.sqlite3 62 | db.sqlite3-journal 63 | 64 | # Flask stuff: 65 | instance/ 66 | .webassets-cache 67 | 68 | # Scrapy stuff: 69 | .scrapy 70 | 71 | # Sphinx documentation 72 | docs/_build/ 73 | 74 | # PyBuilder 75 | target/ 76 | 77 | # Jupyter Notebook 78 | .ipynb_checkpoints 79 | 80 | # IPython 81 | profile_default/ 82 | ipython_config.py 83 | 84 | # pyenv 85 | .python-version 86 | 87 | # pipenv 88 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 89 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 90 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 91 | # install all needed dependencies. 92 | #Pipfile.lock 93 | 94 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow 95 | __pypackages__/ 96 | 97 | # Celery stuff 98 | celerybeat-schedule 99 | celerybeat.pid 100 | 101 | # SageMath parsed files 102 | *.sage.py 103 | 104 | # Environments 105 | .env 106 | .venv 107 | env/ 108 | venv/ 109 | ENV/ 110 | env.bak/ 111 | venv.bak/ 112 | 113 | # Spyder project settings 114 | .spyderproject 115 | .spyproject 116 | 117 | # Rope project settings 118 | .ropeproject 119 | 120 | # mkdocs documentation 121 | /site 122 | 123 | # mypy 124 | .mypy_cache/ 125 | .dmypy.json 126 | dmypy.json 127 | 128 | # Pyre type checker 129 | .pyre/ 130 | 131 | output/* 132 | input/* -------------------------------------------------------------------------------- /kami/run/split_folds.py: -------------------------------------------------------------------------------- 1 | import os 2 | from pathlib import Path 3 | 4 | import hydra 5 | import polars as pl 6 | import yaml 7 | from omegaconf import DictConfig 8 | from sklearn.model_selection import StratifiedKFold 9 | 10 | 11 | def make_fold(cfg: DictConfig, n_folds: int = 5): 12 | series_ids = [ 13 | str(path).split("/")[-1] 14 | for path in (Path(cfg.dir.processed_dir) / "train").glob("*") 15 | ] 16 | 17 | event_df = pl.read_csv(Path(cfg.dir.data_dir) / "train_events.csv").drop_nulls() 18 | q_cut_num = 10 # no_event は含まないので全体では11クラス 19 | event_count_df = event_df.group_by("series_id").count() 20 | event_count_df = event_count_df.select( 21 | pl.col("series_id"), 22 | pl.col("count"), 23 | pl.col("count") 24 | .qcut(q_cut_num, labels=[str(i) for i in range(1, q_cut_num + 1)]) 25 | .alias("class") 26 | .cast(pl.Utf8), 27 | ) 28 | 29 | no_event = [ 30 | si 31 | for si in series_ids 32 | if si not in list(event_count_df.get_column("series_id")) 33 | ] 34 | no_event_df = pl.DataFrame( 35 | { 36 | "series_id": no_event, 37 | "count": [0 for _ in range(len(no_event))], 38 | "class": [str(0) for _ in range(len(no_event))], 39 | } 40 | ) 41 | no_event_df = no_event_df.select( 42 | pl.col("series_id"), 43 | pl.col("count").cast(pl.UInt32), 44 | pl.col("class").cast(pl.Utf8), 45 | ) 46 | 47 | # 二つを結合 48 | all_df = pl.concat([event_count_df, no_event_df]).sort(by="series_id") 49 | 50 | X = all_df.drop("class") 51 | y = all_df.get_column("class") 52 | 53 | skf = StratifiedKFold(n_splits=n_folds) 54 | 55 | os.makedirs(Path(cfg.dir.input_dir) / "folds", exist_ok=True) 56 | for i, (train_index, test_index) in enumerate(skf.split(X, y)): 57 | fold_dict = { 58 | "train_series_ids": list(all_df.get_column("series_id").take(train_index)), 59 | "valid_series_ids": list(all_df.get_column("series_id").take(test_index)), 60 | } 61 | 62 | with open(Path(cfg.dir.input_dir) / f"folds/stratify_fold_{i}.yaml", "w") as wf: 63 | yaml.dump(fold_dict, wf) 64 | 65 | with open( 66 | Path(cfg.dir.kami_dir) / f"run/conf/split/stratify_fold_{i}.yaml", "w" 67 | ) as wf: 68 | yaml.dump(fold_dict, wf) 69 | 70 | 71 | @hydra.main(config_path="conf", config_name="split_folds", version_base="1.2") 72 | def main(cfg: DictConfig): 73 | make_fold(cfg) 74 | 75 | 76 | if __name__ == "__main__": 77 | main() 78 | -------------------------------------------------------------------------------- /shimacos/components/stacking/main_nn.py: -------------------------------------------------------------------------------- 1 | import os 2 | import random 3 | import re 4 | import shutil 5 | from glob import glob 6 | 7 | import components.stacking.runner as runner 8 | import hydra 9 | import numpy as np 10 | import torch 11 | from hydra.utils import instantiate 12 | from omegaconf import DictConfig 13 | 14 | 15 | def set_seed(seed: int): 16 | os.environ["PYTHONHASHSEED"] = str(seed) 17 | random.seed(seed) 18 | np.random.seed(seed) 19 | torch.manual_seed(seed) 20 | torch.cuda.manual_seed(seed) 21 | torch.backends.cudnn.benchmark = True 22 | # torch.backends.cudnn.deterministic = True 23 | # torch.use_deterministic_algorithms(True, warn_only=True) 24 | 25 | 26 | def prepair_dir(config: DictConfig): 27 | """ 28 | Logの保存先を作成 29 | """ 30 | for path in [ 31 | config.store.result_path, 32 | config.store.log_path, 33 | config.store.model_path, 34 | config.store.feature_path, 35 | ]: 36 | if os.path.exists(path) and config.train.warm_start is False and config.data.is_train: 37 | shutil.rmtree(path) 38 | os.makedirs(path, exist_ok=True) 39 | 40 | 41 | def set_up(config: DictConfig): 42 | # Setup 43 | prepair_dir(config) 44 | set_seed(config.train.seed) 45 | os.environ["CUDA_VISIBLE_DEVICES"] = ",".join(config.base.gpu_id) 46 | 47 | 48 | @hydra.main(config_path="../../yamls", config_name="nn.yaml") 49 | def main(config: DictConfig): 50 | os.chdir(config.store.workdir) 51 | set_up(config) 52 | if config.train.warm_start: 53 | checkpoint_path = sorted( 54 | glob(config.store.model_path + "/*epoch*"), 55 | key=lambda path: int(re.split("[=.]", path)[-2]), 56 | )[-1] 57 | print(checkpoint_path) 58 | else: 59 | checkpoint_path = None 60 | 61 | model = getattr(runner, config.runner)(config) 62 | 63 | if config.data.is_train: 64 | trainer = instantiate(config.trainer) 65 | trainer.fit(model) 66 | else: 67 | state_dict = torch.load(checkpoint_path)["state_dict"] 68 | model.load_state_dict(state_dict) 69 | config.trainer.update( 70 | { 71 | "devices": 1, 72 | "logger": None, 73 | "limit_train_batches": 0.0, 74 | "limit_val_batches": 0.0, 75 | "limit_test_batches": 1.0, 76 | "accelerator": None, 77 | } 78 | ) 79 | trainer = instantiate(config.trainer) 80 | trainer.test(model, model.test_dataloader()) 81 | 82 | 83 | if __name__ == "__main__": 84 | main() 85 | -------------------------------------------------------------------------------- /shimacos/components/stacking/sync_batchnorm/batchnorm_reimpl.py: -------------------------------------------------------------------------------- 1 | #! /usr/bin/env python3 2 | # -*- coding: utf-8 -*- 3 | # File : batchnorm_reimpl.py 4 | # Author : acgtyrant 5 | # Date : 11/01/2018 6 | # 7 | # This file is part of Synchronized-BatchNorm-PyTorch. 8 | # https://github.com/vacancy/Synchronized-BatchNorm-PyTorch 9 | # Distributed under MIT License. 10 | 11 | import torch 12 | import torch.nn as nn 13 | import torch.nn.init as init 14 | 15 | __all__ = ["BatchNorm2dReimpl"] 16 | 17 | 18 | class BatchNorm2dReimpl(nn.Module): 19 | """ 20 | A re-implementation of batch normalization, used for testing the numerical 21 | stability. 22 | 23 | Author: acgtyrant 24 | See also: 25 | https://github.com/vacancy/Synchronized-BatchNorm-PyTorch/issues/14 26 | """ 27 | 28 | def __init__(self, num_features, eps=1e-5, momentum=0.1): 29 | super().__init__() 30 | 31 | self.num_features = num_features 32 | self.eps = eps 33 | self.momentum = momentum 34 | self.weight = nn.Parameter(torch.empty(num_features)) 35 | self.bias = nn.Parameter(torch.empty(num_features)) 36 | self.register_buffer("running_mean", torch.zeros(num_features)) 37 | self.register_buffer("running_var", torch.ones(num_features)) 38 | self.reset_parameters() 39 | 40 | def reset_running_stats(self): 41 | self.running_mean.zero_() 42 | self.running_var.fill_(1) 43 | 44 | def reset_parameters(self): 45 | self.reset_running_stats() 46 | init.uniform_(self.weight) 47 | init.zeros_(self.bias) 48 | 49 | def forward(self, input_): 50 | batchsize, channels, height, width = input_.size() 51 | numel = batchsize * height * width 52 | input_ = input_.permute(1, 0, 2, 3).contiguous().view(channels, numel) 53 | sum_ = input_.sum(1) 54 | sum_of_square = input_.pow(2).sum(1) 55 | mean = sum_ / numel 56 | sumvar = sum_of_square - sum_ * mean 57 | 58 | self.running_mean = ( 59 | 1 - self.momentum 60 | ) * self.running_mean + self.momentum * mean.detach() 61 | unbias_var = sumvar / (numel - 1) 62 | self.running_var = ( 63 | 1 - self.momentum 64 | ) * self.running_var + self.momentum * unbias_var.detach() 65 | 66 | bias_var = sumvar / numel 67 | inv_std = 1 / (bias_var + self.eps).pow(0.5) 68 | output = (input_ - mean.unsqueeze(1)) * inv_std.unsqueeze( 69 | 1 70 | ) * self.weight.unsqueeze(1) + self.bias.unsqueeze(1) 71 | 72 | return ( 73 | output.view(channels, batchsize, height, width) 74 | .permute(1, 0, 2, 3) 75 | .contiguous() 76 | ) 77 | 78 | -------------------------------------------------------------------------------- /sakami/src/config.py: -------------------------------------------------------------------------------- 1 | from __future__ import annotations 2 | 3 | import copy 4 | import sys 5 | from pathlib import Path 6 | from typing import Any 7 | 8 | import yaml 9 | 10 | 11 | class Config(dict): 12 | """A dictionary class that allows for attribute-style access of values.""" 13 | 14 | __setattr__ = dict.__setitem__ 15 | 16 | def init(self) -> None: 17 | # update meta_config 18 | exec_file_name = sys.argv[0] 19 | 20 | self.home_dir = Path("./") 21 | self.input_dir = Path("../input") 22 | self.data_dir = self.input_dir / self.competition_name 23 | self.cache_dir = self.input_dir / "cache" 24 | self.output_dir = Path("./output") 25 | self.wandb_dir = Path("./wandb") 26 | 27 | self.run_name = Path(exec_file_name).stem 28 | self.save_dir = Path(self.output_dir) / self.run_name 29 | self.checkpoint_dir = self.save_dir / "checkpoint/" 30 | 31 | # make directories 32 | self.input_dir.mkdir(exist_ok=True) 33 | self.data_dir.mkdir(exist_ok=True) 34 | self.cache_dir.mkdir(exist_ok=True) 35 | self.output_dir.mkdir(exist_ok=True) 36 | self.wandb_dir.mkdir(exist_ok=True) 37 | self.save_dir.mkdir(exist_ok=True) 38 | self.checkpoint_dir.mkdir(exist_ok=True) 39 | 40 | def check_columns(self) -> None: 41 | required_columns = [ 42 | "competition_name", 43 | ] 44 | for required_column in required_columns: 45 | if not hasattr(self, required_column): 46 | raise KeyError(f"Meta config {required_column} must be specified.") 47 | 48 | def __getattr__(self, key: Any) -> Any: 49 | value = super().get(key) 50 | if isinstance(value, dict): 51 | return Config(value) 52 | return value 53 | 54 | def __deepcopy__(self, memo: dict[int, int | list[int]] | None = None) -> Config: 55 | """Prevent errors in the `copy.deepcopy` method. 56 | 57 | References 58 | ---------- 59 | - https://stackoverflow.com/questions/49901590/python-using-copy-deepcopy-on-dotdict 60 | """ 61 | return Config(copy.deepcopy(dict(self), memo=memo)) 62 | 63 | @classmethod 64 | def load(cls, config_path: str) -> Config: 65 | """Load a config file. 66 | 67 | Parameters 68 | ---------- 69 | config_path : str 70 | Path to config file. 71 | 72 | Returns 73 | ------- 74 | Config 75 | Configuration parameters. 76 | """ 77 | if not Path(config_path).exists(): 78 | raise ValueError(f"Configuration file {config_path} does not exist.") 79 | 80 | with open(config_path) as f: 81 | config = cls(yaml.safe_load(f)) 82 | 83 | config.check_columns() 84 | config.init() 85 | return config 86 | -------------------------------------------------------------------------------- /kami/src/utils/common.py: -------------------------------------------------------------------------------- 1 | import math 2 | import os 3 | import random 4 | import sys 5 | import time 6 | from contextlib import contextmanager 7 | 8 | import numpy as np 9 | import pandas as pd 10 | import psutil 11 | 12 | 13 | @contextmanager 14 | def trace(title): 15 | t0 = time.time() 16 | p = psutil.Process(os.getpid()) 17 | m0 = p.memory_info().rss / 2.0**30 18 | yield 19 | m1 = p.memory_info().rss / 2.0**30 20 | delta = m1 - m0 21 | sign = "+" if delta >= 0 else "-" 22 | delta = math.fabs(delta) 23 | print(f"[{m1:.1f}GB({sign}{delta:.1f}GB):{time.time() - t0:.1f}sec] {title} ", file=sys.stderr) 24 | 25 | 26 | def pad_if_needed(x: np.ndarray, max_len: int, pad_value: float = 0.0) -> np.ndarray: 27 | if len(x) == max_len: 28 | return x 29 | num_pad = max_len - len(x) 30 | n_dim = len(x.shape) 31 | pad_widths = [(0, num_pad)] + [(0, 0) for _ in range(n_dim - 1)] 32 | return np.pad(x, pad_width=pad_widths, mode="constant", constant_values=pad_value) 33 | 34 | 35 | def nearest_valid_size(input_size: int, downsample_rate: int) -> int: 36 | """ 37 | (x // hop_length) % 32 == 0 38 | を満たすinput_sizeに最も近いxを返す 39 | """ 40 | 41 | while (input_size // downsample_rate) % 32 != 0: 42 | input_size += 1 43 | assert (input_size // downsample_rate) % 32 == 0 44 | 45 | return input_size 46 | 47 | 48 | def random_crop(pos: int, duration: int, max_end) -> tuple[int, int]: 49 | """Randomly crops with duration length including pos. 50 | However, 0<=start, end<=max_end 51 | """ 52 | start = random.randint(max(0, pos - duration), min(pos, max_end - duration)) 53 | end = start + duration 54 | return start, end 55 | 56 | 57 | def negative_sampling(this_event_df: pd.DataFrame, num_steps: int) -> int: 58 | """negative sampling 59 | 60 | Args: 61 | this_event_df (pd.DataFrame): event df 62 | num_steps (int): number of steps in this series 63 | 64 | Returns: 65 | int: negative sample position 66 | """ 67 | # onsetとwakupを除いた範囲からランダムにサンプリング 68 | positive_positions = set(this_event_df[["onset", "wakeup"]].to_numpy().flatten().tolist()) 69 | negative_positions = list(set(range(num_steps)) - positive_positions) 70 | return random.sample(negative_positions, 1)[0] 71 | 72 | 73 | # ref: https://www.kaggle.com/competitions/dfl-bundesliga-data-shootout/discussion/360236#2004730 74 | def gaussian_kernel(length: int, sigma: int = 3) -> np.ndarray: 75 | x = np.ogrid[-length : length + 1] 76 | h = np.exp(-(x**2) / (2 * sigma * sigma)) # type: ignore 77 | h[h < np.finfo(h.dtype).eps * h.max()] = 0 78 | return h 79 | 80 | 81 | def gaussian_label(label: np.ndarray, offset: int, sigma: int) -> np.ndarray: 82 | num_events = label.shape[1] 83 | for i in range(num_events): 84 | label[:, i] = np.convolve(label[:, i], gaussian_kernel(offset, sigma), mode="same") 85 | 86 | return label 87 | -------------------------------------------------------------------------------- /shimacos/components/factories/sampler_factory.py: -------------------------------------------------------------------------------- 1 | import math 2 | import torch 3 | from torch.utils.data.distributed import DistributedSampler 4 | 5 | 6 | class WeightedDistributedSampler(DistributedSampler): 7 | def __init__(self, dataset, num_replicas=None, rank=None, shuffle=True): 8 | super().__init__(dataset, num_replicas, rank, shuffle) 9 | weights_dict = {0: 1, 1: 1} 10 | weights = ( 11 | dataset.sample_df["pe_present_on_image"] 12 | .apply(lambda x: weights_dict[x]) 13 | .values 14 | ) 15 | self.weights = torch.as_tensor(weights, dtype=torch.double) 16 | 17 | def __iter__(self): 18 | indices = list(range(len(self.dataset))) 19 | # add extra samples to make it evenly divisible 20 | indices += indices[: (self.total_size - len(indices))] 21 | assert len(indices) == self.total_size 22 | 23 | # subsample 24 | indices = indices[self.rank : self.total_size : self.num_replicas] 25 | weights = self.weights[indices] 26 | assert len(indices) == self.num_samples 27 | return ( 28 | indices[i] 29 | for i in torch.multinomial(weights, self.num_samples, replacement=True) 30 | ) 31 | 32 | def __len__(self): 33 | return self.num_samples 34 | 35 | def set_epoch(self, epoch): 36 | self.epoch = epoch 37 | 38 | 39 | class BalancedDistributedSampler(DistributedSampler): 40 | def __init__(self, dataset, num_replicas=None, rank=None, shuffle=True): 41 | self.dataset = dataset 42 | self.num_replicas = num_replicas 43 | self.rank = rank 44 | self.epoch = 0 45 | self.num_samples = int(math.ceil(len(self.dataset) * 1.0 / self.num_replicas)) 46 | self.total_size = self.num_samples * self.num_replicas 47 | self.shuffle = shuffle 48 | 49 | def __iter__(self): 50 | # deterministically shuffle based on epoch 51 | g = torch.Generator() 52 | g.manual_seed(self.epoch) 53 | if self.shuffle: 54 | indices = torch.randperm(len(self.dataset), generator=g).tolist() 55 | else: 56 | indices = list(range(len(self.dataset))) 57 | 58 | # add extra samples to make it evenly divisible 59 | indices += indices[: (self.total_size - len(indices))] 60 | assert len(indices) == self.total_size 61 | 62 | # subsample 63 | indices = indices[self.rank : self.total_size : self.num_replicas] 64 | assert len(indices) == self.num_samples 65 | 66 | return iter(indices) 67 | 68 | def __len__(self): 69 | return self.num_samples 70 | 71 | def set_epoch(self, epoch): 72 | self.epoch = epoch 73 | 74 | 75 | def get_weighted_sampler(dataset): 76 | sampler = WeightedDistributedSampler(dataset=dataset) 77 | return sampler 78 | 79 | 80 | def get_sampler(sampler_class, **params): 81 | print("sampler class:", sampler_class) 82 | f = globals().get(sampler_class) 83 | return f(**params) 84 | -------------------------------------------------------------------------------- /shimacos/components/factories/callback_factory.py: -------------------------------------------------------------------------------- 1 | import os 2 | from glob import glob 3 | 4 | from google.cloud import storage 5 | 6 | from pytorch_lightning.callbacks import ModelCheckpoint, ProgressBar 7 | 8 | 9 | class MyCallback(ModelCheckpoint): 10 | def __init__( 11 | self, 12 | store_config, 13 | monitor="val_loss", 14 | verbose=0, 15 | save_top_k=1, 16 | save_weights_only=False, 17 | mode="auto", 18 | period=1, 19 | ): 20 | super(MyCallback, self).__init__( 21 | store_config.model_path, 22 | monitor, 23 | verbose, 24 | save_top_k, 25 | save_weights_only, 26 | mode, 27 | period, 28 | store_config.model_class, 29 | ) 30 | self.store_config = store_config 31 | 32 | def _save_model(self, filepath): 33 | dirpath = os.path.dirname(filepath) 34 | # make paths 35 | os.makedirs(dirpath, exist_ok=True) 36 | 37 | # delegate the saving to the model 38 | self.save_function(filepath) 39 | if self.store_config.gcs_project is not None: 40 | self.upload_directory() 41 | 42 | def upload_directory(self): 43 | storage_client = storage.Client(self.store_config.gcs_project) 44 | bucket = storage_client.get_bucket(self.store_config.bucket_name) 45 | filenames = glob( 46 | os.path.join(self.store_config.save_path, "**"), recursive=True 47 | ) 48 | for filename in filenames: 49 | if os.path.isdir(filename): 50 | continue 51 | destination_blob_name = os.path.join( 52 | self.store_config.gcs_path, 53 | filename.split(self.store_config.save_path)[-1][1:], 54 | ) 55 | blob = bucket.blob(destination_blob_name) 56 | blob.upload_from_filename(filename) 57 | 58 | 59 | class MyProgressBar(ProgressBar): 60 | def format_num(self, n): 61 | f = "{0:2.3g}".format(n) 62 | f = f.replace("+0", "+") 63 | f = f.replace("-0", "-") 64 | n = str(n) 65 | return f if len(f) < len(n) else n 66 | 67 | def on_batch_end(self, trainer, pl_module): 68 | # super().on_batch_end(trainer, pl_module) 69 | if self.is_enabled and self.train_batch_idx % self.refresh_rate == 0: 70 | self.main_progress_bar.update(self.refresh_rate) 71 | for key, val in trainer.progress_bar_dict.items(): 72 | if not isinstance(val, str): 73 | trainer.progress_bar_dict[key] = self.format_num(val) 74 | self.main_progress_bar.set_postfix(trainer.progress_bar_dict) 75 | 76 | def on_validation_end(self, trainer, pl_module): 77 | super().on_validation_end(trainer, pl_module) 78 | for key, val in trainer.progress_bar_dict.items(): 79 | if not isinstance(val, str): 80 | trainer.progress_bar_dict[key] = self.format_num(val) 81 | self.main_progress_bar.set_postfix(trainer.progress_bar_dict) 82 | self.val_progress_bar.close() 83 | -------------------------------------------------------------------------------- /kami/src/models/feature_extractor/cnn.py: -------------------------------------------------------------------------------- 1 | from typing import Callable, Optional 2 | 3 | import torch 4 | import torch.nn as nn 5 | 6 | 7 | # ref: https://github.com/analokmaus/kaggle-g2net-public/tree/main/models1d_pytorch 8 | class CNNSpectrogram(nn.Module): 9 | def __init__( 10 | self, 11 | in_channels: int = 3, 12 | base_filters: int | tuple = 128, 13 | kernel_sizes: tuple = (32, 16, 4, 2), 14 | stride: int = 4, 15 | sigmoid: bool = False, 16 | output_size: Optional[int] = None, 17 | conv: Callable = nn.Conv1d, 18 | reinit: bool = True, 19 | ): 20 | super().__init__() 21 | self.out_chans = len(kernel_sizes) 22 | self.out_size = output_size 23 | self.sigmoid = sigmoid 24 | if isinstance(base_filters, int): 25 | base_filters = tuple([base_filters]) 26 | self.height = base_filters[-1] 27 | self.spec_conv = nn.ModuleList() 28 | for i in range(self.out_chans): 29 | tmp_block = [ 30 | conv( 31 | in_channels, 32 | base_filters[0], 33 | kernel_size=kernel_sizes[i], 34 | stride=stride, 35 | padding=(kernel_sizes[i] - 1) // 2, 36 | ) 37 | ] 38 | if len(base_filters) > 1: 39 | for j in range(len(base_filters) - 1): 40 | tmp_block = tmp_block + [ 41 | nn.BatchNorm1d(base_filters[j]), 42 | nn.ReLU(inplace=True), 43 | conv( 44 | base_filters[j], 45 | base_filters[j + 1], 46 | kernel_size=kernel_sizes[i], 47 | stride=stride, 48 | padding=(kernel_sizes[i] - 1) // 2, 49 | ), 50 | ] 51 | self.spec_conv.append(nn.Sequential(*tmp_block)) 52 | else: 53 | self.spec_conv.append(tmp_block[0]) 54 | 55 | if self.out_size is not None: 56 | self.pool = nn.AdaptiveAvgPool2d((None, self.out_size)) 57 | 58 | if reinit: 59 | for m in self.modules(): 60 | if isinstance(m, nn.Conv1d): 61 | nn.init.kaiming_normal_(m.weight) 62 | elif isinstance(m, nn.BatchNorm1d): 63 | nn.init.constant_(m.weight, 1) 64 | nn.init.constant_(m.bias, 0) 65 | 66 | def forward(self, x: torch.Tensor) -> torch.Tensor: 67 | """Forward pass of the model. 68 | 69 | Args: 70 | x (_type_): (batch_size, in_channels, time_steps) 71 | 72 | Returns: 73 | _type_: (batch_size, out_chans, height, time_steps) 74 | """ 75 | # x: (batch_size, in_channels, time_steps) 76 | out: list[torch.Tensor] = [] 77 | for i in range(self.out_chans): 78 | out.append(self.spec_conv[i](x)) 79 | img = torch.stack(out, dim=1) # (batch_size, out_chans, height, time_steps) 80 | if self.out_size is not None: 81 | img = self.pool(img) # (batch_size, out_chans, height, out_size) 82 | if self.sigmoid: 83 | img = img.sigmoid() 84 | return img 85 | -------------------------------------------------------------------------------- /shimacos/yamls/stacking.yaml: -------------------------------------------------------------------------------- 1 | defaults: 2 | - feature: "stacking_001" 3 | # - override hydra/job_logging: colorlog 4 | # - override hydra/hydra_logging: colorlog 5 | 6 | workdir: ${store.workdir} 7 | seed: 777 8 | 9 | feature: 10 | sakami_model_names: ["148_gru_scale_factor", "156_gru_transformer_residual"] 11 | kami_model_names: ["exp068_transformer", "exp078_lstm"] 12 | label_col: label 13 | pred_col: ${feature.label_col}_pred 14 | 15 | lgbm: 16 | n_fold: 5 17 | numerical_cols: ${feature.numerical_cols} 18 | cat_cols: ${feature.cat_cols} 19 | label_col: ${feature.label_col} 20 | pred_col: ${feature.pred_col} 21 | early_stopping_rounds: 200 22 | verbose_eval: 100 23 | negative_sampling: False 24 | batch_fit: False 25 | batch_iter: 500 26 | params: 27 | lambda_l1: 0.1 28 | lambda_l2: 0.1 29 | num_leaves: 32 30 | label_gain: 31 | lambdarank_truncation_level: 32 | feature_fraction: 0.8 33 | bagging_fraction: 0.8 34 | bagging_freq: 1 35 | min_child_samples: 10 36 | task: train 37 | boosting_type: gbdt 38 | objective: binary 39 | metric: binary_logloss 40 | max_depth: 8 41 | learning_rate: 0.01 42 | num_thread: -1 43 | max_bin: 256 44 | verbose: -1 45 | device: cpu 46 | scale_pos_weight: 1 47 | seed: ${seed} 48 | num_class: 1 49 | is_unbalance: False 50 | 51 | xgb: 52 | feature_cols: [""] #関数内で書き換える 53 | cat_cols: 54 | label_col: "label" 55 | early_stopping_rounds: 200 56 | verbose_eval: 100 57 | params: 58 | alpha: 0.1 59 | reg_lambda: 0.1 60 | max_leaves: 16 61 | colsample_bytree: 1.0 62 | subsample: 0.8 63 | min_child_weight: 10 64 | booster: gbtree 65 | objective: binary:logistic 66 | eval_metric: auc 67 | max_depth: 6 68 | learning_rate: 0.01 69 | nthread: -1 70 | max_bin: 256 71 | tree_method: gpu_hist 72 | scale_pos_weight: 1 73 | seed: ${seed} 74 | 75 | catboost: 76 | n_fold: 5 77 | numerical_cols: ${feature.numerical_cols} 78 | cat_cols: ${feature.cat_cols} 79 | label_col: ${feature.label_col} 80 | pred_col: ${feature.pred_col} 81 | early_stopping_rounds: 200 82 | verbose_eval: 100 83 | categorical_features_indices: 84 | negative_sampling: False 85 | params: 86 | task_type: GPU 87 | iterations: 100000 88 | loss_function: Logloss 89 | eval_metric: Logloss 90 | custom_metric: Logloss 91 | bootstrap_type: Bernoulli 92 | subsample: 0.8 93 | max_depth: 8 94 | max_ctr_complexity: 4 95 | learning_rate: 0.05 96 | max_bin: 254 97 | verbose: 100 98 | devices: "0" 99 | use_best_model: True 100 | od_type: Iter 101 | od_wait: 100 102 | random_seed: ${seed} 103 | gpu_ram_part: 0.95 104 | allow_writing_files: False 105 | 106 | store: 107 | workdir: ${oc.env:PWD} 108 | model_name: stacking_${feature.version} 109 | root_path: ${store.workdir}/output/${store.model_name} 110 | save_path: ${store.workdir}/output/${store.model_name} 111 | model_path: ${store.workdir}/output/${store.model_name}/model 112 | log_path: ${store.workdir}/output/${store.model_name}/logs 113 | result_path: ${store.workdir}/output/${store.model_name}/result 114 | gcs_path: shimacos/${store.model_name} 115 | save_feature: False 116 | wandb_project: 117 | gcs_project: 118 | bucket_name: 119 | 120 | hydra: 121 | run: 122 | dir: ${store.save_path}/logs 123 | sweep: 124 | dir: ${store.save_path} 125 | subdir: log 126 | job: 127 | chdir: False 128 | -------------------------------------------------------------------------------- /kami/src/models/spec1D.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | from omegaconf import DictConfig 3 | 4 | import torch 5 | import torch.nn as nn 6 | 7 | from src.augmentation.cutmix import Cutmix 8 | from src.augmentation.mixup import Mixup 9 | 10 | from src.models.loss.tolerance import ToleranceLoss 11 | from src.models.loss.tolerance_mse import ToleranceMSELoss 12 | from src.models.loss.bce import BCEWithLogitsLoss 13 | from src.models.loss.tolerance_nonzero import ToleranceNonZeroLoss 14 | 15 | 16 | class Spec1D(nn.Module): 17 | def __init__( 18 | self, 19 | cfg: DictConfig, 20 | feature_extractor: nn.Module, 21 | decoder: nn.Module, 22 | mixup_alpha: float = 0.5, 23 | cutmix_alpha: float = 0.5, 24 | ): 25 | super().__init__() 26 | self.feature_extractor = feature_extractor 27 | self.decoder = decoder 28 | self.channels_fc = nn.Linear(feature_extractor.out_chans, 1) 29 | self.mixup = Mixup(mixup_alpha) 30 | self.cutmix = Cutmix(cutmix_alpha) 31 | self.loss_weight = torch.tensor(cfg.loss.loss_weight) if "loss_weight" in cfg.loss else None 32 | self.label_weight = torch.tensor(cfg.label_weight) if "label_weight" in cfg else None 33 | self.pos_weight = torch.tensor(cfg.pos_weight) if "pos_weight" in cfg else None 34 | 35 | if cfg.loss.name == "tolerance": 36 | self.loss_fn = ToleranceLoss( 37 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 38 | ) 39 | elif cfg.loss.name == "tolerance_mse": 40 | self.loss_fn = ToleranceMSELoss( 41 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 42 | ) 43 | elif cfg.loss.name == "tolerance_nonzero": 44 | self.loss_fn = ToleranceNonZeroLoss( 45 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 46 | ) 47 | else: 48 | self.loss_fn = BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 49 | 50 | def forward( 51 | self, 52 | x: torch.Tensor, 53 | labels: Optional[torch.Tensor] = None, 54 | masks: Optional[torch.Tensor] = None, 55 | do_mixup: bool = False, 56 | do_cutmix: bool = False, 57 | ) -> dict[str, torch.Tensor]: 58 | """Forward pass of the model. 59 | 60 | Args: 61 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 62 | labels (Optional[torch.Tensor], optional): (batch_size, n_timesteps, n_classes) 63 | Returns: 64 | dict[str, torch.Tensor]: logits (batch_size, n_timesteps, n_classes) 65 | """ 66 | x = self.feature_extractor(x) # (batch_size, n_channels, height, n_timesteps) 67 | 68 | if do_mixup and labels is not None: 69 | x, labels = self.mixup(x, labels) 70 | if do_cutmix and labels is not None: 71 | x, labels = self.cutmix(x, labels) 72 | 73 | # pool over n_channels dimension 74 | x = x.transpose(1, 3) # (batch_size, n_timesteps, height, n_channels) 75 | x = self.channels_fc(x) # (batch_size, n_timesteps, height, 1) 76 | x = x.squeeze(-1).transpose(1, 2) # (batch_size, height, n_timesteps) 77 | logits = self.decoder(x) # (batch_size, n_classes, n_timesteps) 78 | 79 | output = {"logits": logits} 80 | if labels is not None: 81 | loss = self.loss_fn(logits, labels, masks) 82 | output["loss"] = loss 83 | 84 | return output 85 | -------------------------------------------------------------------------------- /shimacos/components/stacking/sync_batchnorm/replicate.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | # File : replicate.py 3 | # Author : Jiayuan Mao 4 | # Email : maojiayuan@gmail.com 5 | # Date : 27/01/2018 6 | # 7 | # This file is part of Synchronized-BatchNorm-PyTorch. 8 | # https://github.com/vacancy/Synchronized-BatchNorm-PyTorch 9 | # Distributed under MIT License. 10 | 11 | import functools 12 | 13 | from torch.nn.parallel.data_parallel import DataParallel 14 | 15 | __all__ = [ 16 | 'CallbackContext', 17 | 'execute_replication_callbacks', 18 | 'DataParallelWithCallback', 19 | 'patch_replication_callback' 20 | ] 21 | 22 | 23 | class CallbackContext(object): 24 | pass 25 | 26 | 27 | def execute_replication_callbacks(modules): 28 | """ 29 | Execute an replication callback `__data_parallel_replicate__` on each module created by original replication. 30 | 31 | The callback will be invoked with arguments `__data_parallel_replicate__(ctx, copy_id)` 32 | 33 | Note that, as all modules are isomorphism, we assign each sub-module with a context 34 | (shared among multiple copies of this module on different devices). 35 | Through this context, different copies can share some information. 36 | 37 | We guarantee that the callback on the master copy (the first copy) will be called ahead of calling the callback 38 | of any slave copies. 39 | """ 40 | master_copy = modules[0] 41 | nr_modules = len(list(master_copy.modules())) 42 | ctxs = [CallbackContext() for _ in range(nr_modules)] 43 | 44 | for i, module in enumerate(modules): 45 | for j, m in enumerate(module.modules()): 46 | if hasattr(m, '__data_parallel_replicate__'): 47 | m.__data_parallel_replicate__(ctxs[j], i) 48 | 49 | 50 | class DataParallelWithCallback(DataParallel): 51 | """ 52 | Data Parallel with a replication callback. 53 | 54 | An replication callback `__data_parallel_replicate__` of each module will be invoked after being created by 55 | original `replicate` function. 56 | The callback will be invoked with arguments `__data_parallel_replicate__(ctx, copy_id)` 57 | 58 | Examples: 59 | > sync_bn = SynchronizedBatchNorm1d(10, eps=1e-5, affine=False) 60 | > sync_bn = DataParallelWithCallback(sync_bn, device_ids=[0, 1]) 61 | # sync_bn.__data_parallel_replicate__ will be invoked. 62 | """ 63 | 64 | def replicate(self, module, device_ids): 65 | modules = super(DataParallelWithCallback, self).replicate(module, device_ids) 66 | execute_replication_callbacks(modules) 67 | return modules 68 | 69 | 70 | def patch_replication_callback(data_parallel): 71 | """ 72 | Monkey-patch an existing `DataParallel` object. Add the replication callback. 73 | Useful when you have customized `DataParallel` implementation. 74 | 75 | Examples: 76 | > sync_bn = SynchronizedBatchNorm1d(10, eps=1e-5, affine=False) 77 | > sync_bn = DataParallel(sync_bn, device_ids=[0, 1]) 78 | > patch_replication_callback(sync_bn) 79 | # this is equivalent to 80 | > sync_bn = SynchronizedBatchNorm1d(10, eps=1e-5, affine=False) 81 | > sync_bn = DataParallelWithCallback(sync_bn, device_ids=[0, 1]) 82 | """ 83 | 84 | assert isinstance(data_parallel, DataParallel) 85 | 86 | old_replicate = data_parallel.replicate 87 | 88 | @functools.wraps(old_replicate) 89 | def new_replicate(module, device_ids): 90 | modules = old_replicate(module, device_ids) 91 | execute_replication_callbacks(modules) 92 | return modules 93 | 94 | data_parallel.replicate = new_replicate 95 | -------------------------------------------------------------------------------- /sakami/src/utils.py: -------------------------------------------------------------------------------- 1 | import functools 2 | import gc 3 | import os 4 | import random 5 | import time 6 | from contextlib import ContextDecorator 7 | from typing import Any, Callable, Type 8 | 9 | import numpy as np 10 | import torch 11 | from loguru import logger 12 | 13 | 14 | def set_seed(seed: int = 0) -> None: 15 | random.seed(seed) 16 | os.environ["PYTHONHASHSEED"] = str(seed) 17 | np.random.seed(seed) 18 | torch.manual_seed(seed) 19 | torch.cuda.manual_seed(seed) 20 | torch.backends.cudnn.deterministic = True 21 | torch.backends.cudnn.benchmark = False 22 | 23 | 24 | def freeze(cls) -> Type: 25 | """Decorator function for fixing class variables. 26 | 27 | Examples 28 | -------- 29 | 30 | >>> @freeze 31 | >>> class config: 32 | >>> x = 10 33 | >>> y = 20 34 | 35 | >>> config.x = 30 36 | ValueError: Cannot overwrite config. 37 | """ 38 | 39 | class _Const(type): 40 | """Metaclass of the configuration class. 41 | 42 | Examples 43 | -------- 44 | 45 | >>> class config(metaclass=_Const): 46 | >>> x = 10 47 | >>> y = 20 48 | 49 | >>> config.x = 30 50 | ValueError: Cannot overwrite config. 51 | 52 | References 53 | ---------- 54 | - https://cream-worker.blog.jp/archives/1077207909.html 55 | """ 56 | 57 | def __setattr__(self, name, value): 58 | raise ValueError("Cannot overwrite config.") 59 | 60 | class frozen_cls(cls, metaclass=_Const): 61 | pass 62 | 63 | return frozen_cls 64 | 65 | 66 | class timer(ContextDecorator): 67 | """Context-manager that logs elapsed time of a process. 68 | Also functions as a decorator. (Make sure to instantiate with parenthesis.) 69 | 70 | Paramters 71 | --------- 72 | message : str 73 | The displayed message. 74 | 75 | Examples 76 | -------- 77 | - Usage as a context-manager 78 | 79 | >>> with timer('read csv'): 80 | >>> train_df = pd.read_csv(TRAIN_PATH) 81 | [read csv] start. 82 | [read csv] done in 0.1 min. 83 | 84 | - Usage as a decorator 85 | 86 | >>> @timer() 87 | >>> def read_csv(): 88 | >>> train_df = pd.read_csv(TRAIN_PATH) 89 | >>> return train_df 90 | >>> 91 | >>> train_df = read_csv() 92 | [read_csv] start. 93 | [read_csv] done in 0.1 min. 94 | """ 95 | 96 | def __init__(self, message: str | None = None) -> None: 97 | self.message = message 98 | 99 | def __call__(self, function: Callable) -> Callable: 100 | if self.message is None: 101 | self.message = function.__name__ 102 | return super().__call__(function) 103 | 104 | def __enter__(self) -> None: 105 | self.start_time = time.time() 106 | logger.opt(colors=True).info(f"[{self.message}] start.") 107 | 108 | def __exit__(self, exc_type: Any, exc_value: Any, traceback: Any) -> None: 109 | if exc_type is None: 110 | elapsed_time = time.time() - self.start_time 111 | logger.opt(colors=True).info(f"[{self.message}] done in {elapsed_time / 60:.1f} min.") 112 | 113 | 114 | def clear_memory(function: Callable) -> Callable: 115 | """Decorator function for clearing memory cache.""" 116 | 117 | @functools.wraps(function) 118 | def _clear_memory(*args, **kwargs): 119 | function(*args, **kwargs) 120 | gc.collect() 121 | torch.cuda.empty_cache() 122 | 123 | return _clear_memory 124 | -------------------------------------------------------------------------------- /shimacos/components/factories/scheduler_factory.py: -------------------------------------------------------------------------------- 1 | from torch.optim.lr_scheduler import _LRScheduler 2 | from torch.optim.lr_scheduler import ReduceLROnPlateau 3 | 4 | 5 | class GradualWarmupScheduler(_LRScheduler): 6 | """Gradually warm-up(increasing) learning rate in optimizer. 7 | Proposed in 'Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour'. 8 | Args: 9 | optimizer (Optimizer): Wrapped optimizer. 10 | multiplier: target learning rate = base lr * multiplier 11 | total_epoch: target learning rate is reached at total_epoch, gradually 12 | after_scheduler: after target_epoch, use this scheduler(eg. ReduceLROnPlateau) 13 | """ 14 | 15 | def __init__(self, optimizer, multiplier, total_epoch, after_scheduler=None): 16 | self.multiplier = multiplier 17 | if self.multiplier <= 1.0: 18 | raise ValueError("multiplier should be greater than 1.") 19 | self.total_epoch = total_epoch 20 | self.after_scheduler = after_scheduler 21 | self.finished = False 22 | super().__init__(optimizer) 23 | 24 | def get_lr(self): 25 | if self.last_epoch > self.total_epoch: 26 | if self.after_scheduler: 27 | if not self.finished: 28 | self.after_scheduler.base_lrs = [ 29 | base_lr * self.multiplier for base_lr in self.base_lrs 30 | ] 31 | self.finished = True 32 | return self.after_scheduler.get_lr() 33 | return [base_lr * self.multiplier for base_lr in self.base_lrs] 34 | 35 | return [ 36 | base_lr 37 | * ((self.multiplier - 1.0) * self.last_epoch / self.total_epoch + 1.0) 38 | for base_lr in self.base_lrs 39 | ] 40 | 41 | def step_ReduceLROnPlateau(self, metrics, epoch=None): 42 | if epoch is None: 43 | epoch = self.last_epoch + 1 44 | self.last_epoch = ( 45 | epoch if epoch != 0 else 1 46 | ) # ReduceLROnPlateau is called at the end of epoch, whereas others are called at beginning 47 | if self.last_epoch <= self.total_epoch: 48 | warmup_lr = [ 49 | base_lr 50 | * ((self.multiplier - 1.0) * self.last_epoch / self.total_epoch + 1.0) 51 | for base_lr in self.base_lrs 52 | ] 53 | for param_group, lr in zip(self.optimizer.param_groups, warmup_lr): 54 | param_group["lr"] = lr 55 | else: 56 | if epoch is None: 57 | self.after_scheduler.step(metrics, None) 58 | else: 59 | self.after_scheduler.step(metrics, epoch - self.total_epoch) 60 | 61 | def step(self, epoch=None, metrics=None): 62 | if type(self.after_scheduler) != ReduceLROnPlateau: 63 | if self.finished and self.after_scheduler: 64 | if epoch is None: 65 | self.after_scheduler.step(None) 66 | else: 67 | self.after_scheduler.step(epoch - self.total_epoch) 68 | else: 69 | return super(GradualWarmupScheduler, self).step(epoch) 70 | else: 71 | self.step_ReduceLROnPlateau(metrics, epoch) 72 | 73 | 74 | def get_warmup_scheduler(optimizer, multiplier, total_epoch, after_scheduler): 75 | scheduler = GradualWarmupScheduler( 76 | optimizer, multiplier, total_epoch, after_scheduler 77 | ) 78 | return scheduler 79 | 80 | 81 | def get_plateau_scheduler(optimizer): 82 | scheduler = ReduceLROnPlateau(optimizer, mode="max", factor=0.5, patience=2) 83 | return scheduler 84 | 85 | 86 | def get_scheduler(scheduler_class, **params): 87 | print("scheduler class:", scheduler_class) 88 | f = globals().get(scheduler_class) 89 | return f(**params) 90 | -------------------------------------------------------------------------------- /kami/src/utils/score.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import polars as pl 3 | from tqdm.auto import tqdm 4 | 5 | from src.utils.metrics import event_detection_ap 6 | from src.utils.post_process import post_process_for_seg, post_process_for_seg_group_by_day 7 | 8 | 9 | def score_group_by_day(val_event_df: pl.DataFrame, keys: list[str], preds: np.ndarray, val_df: pl.DataFrame) -> float: 10 | """ 11 | 日毎に最大値のeventを検出し、それをsubmissionとしてスコアリングする 12 | 13 | Args: 14 | val_event_df (pl.DataFrame): Ground truth 15 | keys (list[str]): 予測したchunkのkey({series_id}_{chunk_id}) 16 | preds (np.ndarray): (chunk_num, duration, 2) 17 | val_df (pl.DataFrame): sequence 18 | """ 19 | submission_df = post_process_for_seg_group_by_day(keys, preds, val_df) 20 | score = event_detection_ap( 21 | val_event_df.to_pandas(), 22 | submission_df.to_pandas(), 23 | ) 24 | return score 25 | 26 | 27 | def score_ternary_search_distance( 28 | val_event_df: pl.DataFrame, keys: list[str], preds: np.ndarray, score_th: float = 0.005 29 | ) -> [float, float]: 30 | """ 31 | post_process_for_seg のパラメータdistanceを ternary searchで探索する 32 | """ 33 | l = 5 34 | r = 100 35 | 36 | cnt = 0 37 | best_score = 0.0 38 | best_distance = 0 39 | 40 | for cnt in tqdm(range(30)): 41 | if r - l < 1: 42 | break 43 | m1 = int(l + (r - l) / 3) 44 | m2 = int(r - (r - l) / 3) 45 | score1 = event_detection_ap( 46 | val_event_df.to_pandas(), 47 | post_process_for_seg( 48 | keys=keys, 49 | preds=preds, 50 | score_th=score_th, 51 | distance=m1, 52 | ).to_pandas(), 53 | ) 54 | score2 = event_detection_ap( 55 | val_event_df.to_pandas(), 56 | post_process_for_seg( 57 | keys=keys, 58 | preds=preds, 59 | score_th=score_th, 60 | distance=m2, 61 | ).to_pandas(), 62 | ) 63 | 64 | if score1 >= score2: 65 | r = m2 66 | best_score = score1 67 | best_distance = m1 68 | 69 | else: 70 | l = m1 71 | best_score = score2 72 | best_distance = m2 73 | 74 | tqdm.write(f"score1(m1): {score1:.5f}({m1:.5f}), score2(m2): {score2:.5f}({m2:.5f}), l: {l:.5f}, r: {r:.5f}") 75 | 76 | if abs(m2 - m1) <= 2: 77 | break 78 | 79 | return best_score, best_distance 80 | 81 | 82 | def score_ternary_search_th( 83 | val_event_df: pl.DataFrame, keys: list[str], preds: np.ndarray, distance: int = 5000 84 | ) -> [float, float]: 85 | """ 86 | post_process_for_seg のパラメータ score_th を ternary searchで探索する 87 | """ 88 | l = 0.0 89 | r = 1.0 90 | 91 | cnt = 0 92 | best_score = 0.0 93 | best_th = 0.0 94 | 95 | for cnt in tqdm(range(30)): 96 | if r - l < 0.01: 97 | break 98 | m1 = l + (r - l) / 3 99 | m2 = r - (r - l) / 3 100 | score1 = event_detection_ap( 101 | val_event_df.to_pandas(), 102 | post_process_for_seg( 103 | keys=keys, 104 | preds=preds, 105 | score_th=m1, 106 | distance=distance, 107 | ).to_pandas(), 108 | ) 109 | score2 = event_detection_ap( 110 | val_event_df.to_pandas(), 111 | post_process_for_seg( 112 | keys=keys, 113 | preds=preds, 114 | score_th=m2, 115 | distance=distance, 116 | ).to_pandas(), 117 | ) 118 | if score1 >= score2: 119 | r = m2 120 | best_score = score1 121 | best_th = m1 122 | else: 123 | l = m1 124 | best_score = score2 125 | best_th = m2 126 | 127 | tqdm.write(f"score1(m1): {score1:.5f}({m1:.5f}), score2(m2): {score2:.5f}({m2:.5f}), l: {l:.5f}, r: {r:.5f}") 128 | 129 | return best_score, best_th 130 | -------------------------------------------------------------------------------- /kami/.gitignore: -------------------------------------------------------------------------------- 1 | /input 2 | /output 3 | /processed 4 | /wandb 5 | /.keras 6 | /.config 7 | /.py3 8 | /.nv 9 | /.local 10 | /.ipython 11 | /.virtual_documents 12 | /.jupyter 13 | /.cupy 14 | .netrc 15 | /notebook/wandb 16 | /data 17 | /db 18 | .bash_history 19 | *.parquet 20 | # Byte-compiled / optimized / DLL files 21 | __pycache__/ 22 | *.py[cod] 23 | *$py.class 24 | 25 | # C extensions 26 | *.so 27 | 28 | # Distribution / packaging 29 | .Python 30 | build/ 31 | develop-eggs/ 32 | dist/ 33 | downloads/ 34 | eggs/ 35 | .eggs/ 36 | lib/ 37 | lib64/ 38 | parts/ 39 | sdist/ 40 | var/ 41 | wheels/ 42 | share/python-wheels/ 43 | *.egg-info/ 44 | .installed.cfg 45 | *.egg 46 | MANIFEST 47 | 48 | # PyInstaller 49 | # Usually these files are written by a python script from a template 50 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 51 | *.manifest 52 | *.spec 53 | 54 | # Installer logs 55 | pip-log.txt 56 | pip-delete-this-directory.txt 57 | 58 | # Unit test / coverage reports 59 | htmlcov/ 60 | .tox/ 61 | .nox/ 62 | .coverage 63 | .coverage.* 64 | .cache 65 | nosetests.xml 66 | coverage.xml 67 | *.cover 68 | *.py,cover 69 | .hypothesis/ 70 | .pytest_cache/ 71 | cover/ 72 | 73 | # Translations 74 | *.mo 75 | *.pot 76 | 77 | # Django stuff: 78 | *.log 79 | local_settings.py 80 | db.sqlite3 81 | db.sqlite3-journal 82 | 83 | # Flask stuff: 84 | instance/ 85 | .webassets-cache 86 | 87 | # Scrapy stuff: 88 | .scrapy 89 | 90 | # Sphinx documentation 91 | docs/_build/ 92 | 93 | # PyBuilder 94 | .pybuilder/ 95 | target/ 96 | 97 | # Jupyter Notebook 98 | .ipynb_checkpoints 99 | 100 | # IPython 101 | profile_default/ 102 | ipython_config.py 103 | 104 | # pyenv 105 | # For a library or package, you might want to ignore these files since the code is 106 | # intended to run in multiple environments; otherwise, check them in: 107 | # .python-version 108 | 109 | # pipenv 110 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 111 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 112 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 113 | # install all needed dependencies. 114 | #Pipfile.lock 115 | 116 | # poetry 117 | # Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control. 118 | # This is especially recommended for binary packages to ensure reproducibility, and is more 119 | # commonly ignored for libraries. 120 | # https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control 121 | #poetry.lock 122 | 123 | # pdm 124 | # Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control. 125 | #pdm.lock 126 | # pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it 127 | # in version control. 128 | # https://pdm.fming.dev/#use-with-ide 129 | .pdm.toml 130 | 131 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm 132 | __pypackages__/ 133 | 134 | # Celery stuff 135 | celerybeat-schedule 136 | celerybeat.pid 137 | 138 | # SageMath parsed files 139 | *.sage.py 140 | 141 | # Environments 142 | .env 143 | .venv 144 | env/ 145 | venv/ 146 | ENV/ 147 | env.bak/ 148 | venv.bak/ 149 | 150 | # Spyder project settings 151 | .spyderproject 152 | .spyproject 153 | 154 | # Rope project settings 155 | .ropeproject 156 | 157 | # mkdocs documentation 158 | /site 159 | 160 | # mypy 161 | .mypy_cache/ 162 | .dmypy.json 163 | dmypy.json 164 | 165 | # Pyre type checker 166 | .pyre/ 167 | 168 | # pytype static type analyzer 169 | .pytype/ 170 | 171 | # Cython debug symbols 172 | cython_debug/ 173 | 174 | # PyCharm 175 | # JetBrains specific template is maintained in a separate JetBrains.gitignore that can 176 | # be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore 177 | # and can be added to the global gitignore or merged into this file. For a more nuclear 178 | # option (not recommended) you can uncomment the following to ignore the entire idea folder. 179 | #.idea/ 180 | -------------------------------------------------------------------------------- /shimacos/yamls/feature/stacking_009.yaml: -------------------------------------------------------------------------------- 1 | cat_cols: 2 | - time_idx 3 | - hour 4 | - minute 5 | - minute_mod15 6 | - weekday 7 | label_col: label 8 | numerical_cols: 9 | - pred_onset_148_gru_scale_factor 10 | - pred_onset_156_gru_transformer_residual 11 | - pred_onset_163_gru_sleep_target 12 | - pred_onset_exp068_transformer 13 | - pred_onset_exp078_lstm 14 | - pred_wakeup_148_gru_scale_factor 15 | - pred_wakeup_156_gru_transformer_residual 16 | - pred_wakeup_163_gru_sleep_target 17 | - pred_wakeup_exp068_transformer 18 | - pred_wakeup_exp078_lstm 19 | - pred_sleep_exp068_transformer 20 | - pred_sleep_exp078_lstm 21 | - pred_sleep_163_gru_sleep_target 22 | - anglez 23 | - enmo 24 | - prediction_onset 25 | - prediction_onset_min 26 | - prediction_onset_max 27 | - prediction_wakeup 28 | - prediction_wakeup_min 29 | - prediction_wakeup_max 30 | - periodicity 31 | - activity_count 32 | - lid 33 | - enmo_3_rolling_mean 34 | - enmo_3_rolling_std 35 | - enmo_3_rolling_max 36 | - anglez_3_rolling_mean 37 | - anglez_3_rolling_std 38 | - anglez_3_rolling_max 39 | - lid_3_rolling_mean 40 | - lid_3_rolling_std 41 | - lid_3_rolling_max 42 | - enmo_diff_prev3_rolling_mean 43 | - anglez_diff_prev3_rolling_mean 44 | - lid_diff_prev3_rolling_mean 45 | - enmo_diff_lead3_rolling_mean 46 | - anglez_diff_lead3_rolling_mean 47 | - lid_diff_lead3_rolling_mean 48 | - enmo_diff_lag 49 | - anglez_diff_lag 50 | - lid_diff_lag 51 | - pred_onset_148_gru_scale_factor_diff_prev3_rolling_mean 52 | - pred_onset_156_gru_transformer_residual_diff_prev3_rolling_mean 53 | - pred_onset_163_gru_sleep_target_diff_prev3_rolling_mean 54 | - pred_onset_exp068_transformer_diff_prev3_rolling_mean 55 | - pred_onset_exp078_lstm_diff_prev3_rolling_mean 56 | - pred_wakeup_148_gru_scale_factor_diff_prev3_rolling_mean 57 | - pred_wakeup_156_gru_transformer_residual_diff_prev3_rolling_mean 58 | - pred_wakeup_163_gru_sleep_target_diff_prev3_rolling_mean 59 | - pred_wakeup_exp068_transformer_diff_prev3_rolling_mean 60 | - pred_wakeup_exp078_lstm_diff_prev3_rolling_mean 61 | - pred_onset_148_gru_scale_factor_diff_prev5_rolling_mean 62 | - pred_onset_156_gru_transformer_residual_diff_prev5_rolling_mean 63 | - pred_onset_163_gru_sleep_target_diff_prev5_rolling_mean 64 | - pred_onset_exp068_transformer_diff_prev5_rolling_mean 65 | - pred_onset_exp078_lstm_diff_prev5_rolling_mean 66 | - pred_wakeup_148_gru_scale_factor_diff_prev5_rolling_mean 67 | - pred_wakeup_156_gru_transformer_residual_diff_prev5_rolling_mean 68 | - pred_wakeup_163_gru_sleep_target_diff_prev5_rolling_mean 69 | - pred_wakeup_exp068_transformer_diff_prev5_rolling_mean 70 | - pred_wakeup_exp078_lstm_diff_prev5_rolling_mean 71 | - pred_onset_148_gru_scale_factor_diff_lead3_rolling_mean 72 | - pred_onset_156_gru_transformer_residual_diff_lead3_rolling_mean 73 | - pred_onset_163_gru_sleep_target_diff_lead3_rolling_mean 74 | - pred_onset_exp068_transformer_diff_lead3_rolling_mean 75 | - pred_onset_exp078_lstm_diff_lead3_rolling_mean 76 | - pred_wakeup_148_gru_scale_factor_diff_lead3_rolling_mean 77 | - pred_wakeup_156_gru_transformer_residual_diff_lead3_rolling_mean 78 | - pred_wakeup_163_gru_sleep_target_diff_lead3_rolling_mean 79 | - pred_wakeup_exp068_transformer_diff_lead3_rolling_mean 80 | - pred_wakeup_exp078_lstm_diff_lead3_rolling_mean 81 | - pred_onset_148_gru_scale_factor_diff_lead5_rolling_mean 82 | - pred_onset_156_gru_transformer_residual_diff_lead5_rolling_mean 83 | - pred_onset_163_gru_sleep_target_diff_lead5_rolling_mean 84 | - pred_onset_exp068_transformer_diff_lead5_rolling_mean 85 | - pred_onset_exp078_lstm_diff_lead5_rolling_mean 86 | - pred_wakeup_148_gru_scale_factor_diff_lead5_rolling_mean 87 | - pred_wakeup_156_gru_transformer_residual_diff_lead5_rolling_mean 88 | - pred_wakeup_163_gru_sleep_target_diff_lead5_rolling_mean 89 | - pred_wakeup_exp068_transformer_diff_lead5_rolling_mean 90 | - pred_wakeup_exp078_lstm_diff_lead5_rolling_mean 91 | - pred_onset_148_gru_scale_factor_diff_lag 92 | - pred_onset_156_gru_transformer_residual_diff_lag 93 | - pred_onset_163_gru_sleep_target_diff_lag 94 | - pred_onset_exp068_transformer_diff_lag 95 | - pred_onset_exp078_lstm_diff_lag 96 | - pred_wakeup_148_gru_scale_factor_diff_lag 97 | - pred_wakeup_156_gru_transformer_residual_diff_lag 98 | - pred_wakeup_163_gru_sleep_target_diff_lag 99 | - pred_wakeup_exp068_transformer_diff_lag 100 | - pred_wakeup_exp078_lstm_diff_lag 101 | pred_col: label_pred 102 | version: 009 103 | -------------------------------------------------------------------------------- /shimacos/components/factories/optimizer_factory.py: -------------------------------------------------------------------------------- 1 | import math 2 | import torch 3 | from torch.optim.optimizer import Optimizer, required 4 | from torch.optim import Adam, SGD, AdamW 5 | 6 | 7 | class RAdam(Optimizer): 8 | def __init__( 9 | self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=1e-4 10 | ): 11 | defaults = dict(lr=lr, betas=betas, eps=eps, weight_decay=weight_decay) 12 | self.buffer = [[None, None, None] for ind in range(10)] 13 | super(RAdam, self).__init__(params, defaults) 14 | 15 | def __setstate__(self, state): 16 | super(RAdam, self).__setstate__(state) 17 | 18 | def step(self, closure=None): 19 | 20 | loss = None 21 | if closure is not None: 22 | loss = closure() 23 | 24 | for group in self.param_groups: 25 | 26 | for p in group["params"]: 27 | if p.grad is None: 28 | continue 29 | grad = p.grad.data.float() 30 | if grad.is_sparse: 31 | raise RuntimeError("RAdam does not support sparse gradients") 32 | 33 | p_data_fp32 = p.data.float() 34 | 35 | state = self.state[p] 36 | 37 | if len(state) == 0: 38 | state["step"] = 0 39 | state["exp_avg"] = torch.zeros_like(p_data_fp32) 40 | state["exp_avg_sq"] = torch.zeros_like(p_data_fp32) 41 | else: 42 | state["exp_avg"] = state["exp_avg"].type_as(p_data_fp32) 43 | state["exp_avg_sq"] = state["exp_avg_sq"].type_as(p_data_fp32) 44 | 45 | exp_avg, exp_avg_sq = state["exp_avg"], state["exp_avg_sq"] 46 | beta1, beta2 = group["betas"] 47 | 48 | exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad) 49 | exp_avg.mul_(beta1).add_(1 - beta1, grad) 50 | 51 | state["step"] += 1 52 | buffered = self.buffer[int(state["step"] % 10)] 53 | if state["step"] == buffered[0]: 54 | N_sma, step_size = buffered[1], buffered[2] 55 | else: 56 | buffered[0] = state["step"] 57 | beta2_t = beta2 ** state["step"] 58 | N_sma_max = 2 / (1 - beta2) - 1 59 | N_sma = N_sma_max - 2 * state["step"] * beta2_t / (1 - beta2_t) 60 | buffered[1] = N_sma 61 | 62 | # more conservative since it's an approximated value 63 | if N_sma >= 5: 64 | step_size = math.sqrt( 65 | (1 - beta2_t) 66 | * (N_sma - 4) 67 | / (N_sma_max - 4) 68 | * (N_sma - 2) 69 | / N_sma 70 | * N_sma_max 71 | / (N_sma_max - 2) 72 | ) / (1 - beta1 ** state["step"]) 73 | else: 74 | step_size = 1.0 / (1 - beta1 ** state["step"]) 75 | buffered[2] = step_size 76 | 77 | if group["weight_decay"] != 0: 78 | p_data_fp32.add_(-group["weight_decay"] * group["lr"], p_data_fp32) 79 | 80 | # more conservative since it's an approximated value 81 | if N_sma >= 5: 82 | denom = exp_avg_sq.sqrt().add_(group["eps"]) 83 | p_data_fp32.addcdiv_(-step_size * group["lr"], exp_avg, denom) 84 | else: 85 | p_data_fp32.add_(-step_size * group["lr"], exp_avg) 86 | 87 | p.data.copy_(p_data_fp32) 88 | 89 | return loss 90 | 91 | 92 | def get_adam(params, lr): 93 | optimizer = Adam(params=params, lr=lr) 94 | return optimizer 95 | 96 | 97 | def get_adamw(params, lr): 98 | optimizer = AdamW(params=params, lr=lr) 99 | return optimizer 100 | 101 | 102 | def get_sgd(params, lr): 103 | optimizer = SGD(params=params, lr=lr, momentum=0.9, weight_decay=1e-4) 104 | return optimizer 105 | 106 | 107 | def get_radam(params, lr): 108 | optimizer = RAdam(params=params, lr=lr) 109 | return optimizer 110 | 111 | 112 | def get_optimizer(opt_class, **params): 113 | print("optimizer class:", opt_class) 114 | f = globals().get(opt_class) 115 | return f(**params) 116 | -------------------------------------------------------------------------------- /kami/src/models/spec2Dcnn.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | from omegaconf import DictConfig 3 | 4 | import segmentation_models_pytorch as smp 5 | import torch 6 | import torch.nn as nn 7 | 8 | from src.augmentation.cutmix import Cutmix 9 | from src.augmentation.mixup import Mixup 10 | 11 | from src.models.loss.tolerance import ToleranceLoss 12 | from src.models.loss.tolerance_mse import ToleranceMSELoss 13 | from src.models.loss.bce import BCEWithLogitsLoss 14 | from src.models.loss.tolerance_nonzero import ToleranceNonZeroLoss 15 | from src.models.loss.focal import FocalLoss 16 | from src.models.loss.focal_bce import FocalBCELoss 17 | 18 | 19 | class Spec2DCNN(nn.Module): 20 | def __init__( 21 | self, 22 | cfg: DictConfig, 23 | feature_extractor: nn.Module, 24 | decoder: nn.Module, 25 | encoder_name: str, 26 | in_channels: int, 27 | encoder_weights: Optional[str] = None, 28 | mixup_alpha: float = 0.5, 29 | cutmix_alpha: float = 0.5, 30 | ): 31 | super().__init__() 32 | self.cfg = cfg 33 | self.feature_extractor = feature_extractor 34 | self.encoder = smp.Unet( 35 | encoder_name=encoder_name, 36 | encoder_weights=encoder_weights, 37 | in_channels=in_channels, 38 | classes=1, 39 | ) 40 | self.decoder = decoder 41 | self.mixup = Mixup(mixup_alpha) 42 | self.cutmix = Cutmix(cutmix_alpha) 43 | self.loss_weight = torch.tensor(cfg.loss.loss_weight) if "loss_weight" in cfg.loss else None 44 | self.label_weight = torch.tensor(cfg.label_weight) if "label_weight" in cfg else None 45 | self.pos_weight = torch.tensor(cfg.pos_weight) if "pos_weight" in cfg else None 46 | self.loss_fn = None 47 | self.update_loss_fn() 48 | 49 | def update_loss_fn(self, sleep_decay: float = 1.0) -> None: 50 | self.label_weight[0] = self.label_weight[0] * sleep_decay 51 | if self.cfg.loss.name == "tolerance": 52 | self.loss_fn = ToleranceLoss( 53 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 54 | ) 55 | elif self.cfg.loss.name == "tolerance_mse": 56 | self.loss_fn = ToleranceMSELoss( 57 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 58 | ) 59 | elif self.cfg.loss.name == "tolerance_nonzero": 60 | self.loss_fn = ToleranceNonZeroLoss( 61 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 62 | ) 63 | elif self.cfg.loss.name == "focal": 64 | self.loss_fn = FocalLoss( 65 | alpha=self.cfg.loss.alpha, 66 | gamma=self.cfg.loss.gamma, 67 | ) 68 | elif self.cfg.loss.name == "focal_bce": 69 | self.loss_fn = FocalBCELoss( 70 | alpha=self.cfg.loss.alpha, 71 | gamma=self.cfg.loss.gamma, 72 | weight=torch.tensor(self.cfg.loss.weight), 73 | ) 74 | else: 75 | self.loss_fn = BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 76 | 77 | self.loss_fn = self.loss_fn.cuda() 78 | 79 | def forward( 80 | self, 81 | x: torch.Tensor, 82 | labels: Optional[torch.Tensor] = None, 83 | masks: Optional[torch.Tensor] = None, 84 | do_mixup: bool = False, 85 | do_cutmix: bool = False, 86 | ) -> dict[str, torch.Tensor]: 87 | """Forward pass of the model. 88 | 89 | Args: 90 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 91 | labels (Optional[torch.Tensor], optional): (batch_size, n_timesteps, n_classes) 92 | Returns: 93 | dict[str, torch.Tensor]: logits (batch_size, n_timesteps, n_classes) 94 | """ 95 | x = self.feature_extractor(x) # (batch_size, n_channels, height, n_timesteps) 96 | 97 | if do_mixup and labels is not None: 98 | x, labels = self.mixup(x, labels) 99 | if do_cutmix and labels is not None: 100 | x, labels = self.cutmix(x, labels) 101 | 102 | x = self.encoder(x).squeeze(1) # (batch_size, height, n_timesteps) 103 | logits = self.decoder(x) # (batch_size, n_timesteps, n_classes) 104 | 105 | output = {"logits": logits} 106 | if labels is not None: 107 | loss = self.loss_fn(logits, labels, masks) 108 | output["loss"] = loss 109 | 110 | return output 111 | -------------------------------------------------------------------------------- /kami/src/models/spec2DcnnOverlap.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | from omegaconf import DictConfig 3 | 4 | import segmentation_models_pytorch as smp 5 | import torch 6 | import torch.nn as nn 7 | 8 | from src.augmentation.cutmix import Cutmix 9 | from src.augmentation.mixup import Mixup 10 | 11 | from src.models.loss.tolerance import ToleranceLoss 12 | from src.models.loss.tolerance_mse import ToleranceMSELoss 13 | from src.models.loss.bce import BCEWithLogitsLoss 14 | from src.models.loss.tolerance_nonzero import ToleranceNonZeroLoss 15 | from src.models.loss.focal import FocalLoss 16 | from src.models.loss.focal_bce import FocalBCELoss 17 | 18 | 19 | class Spec2DCNNOverlap(nn.Module): 20 | def __init__( 21 | self, 22 | cfg: DictConfig, 23 | feature_extractor: nn.Module, 24 | decoder: nn.Module, 25 | encoder_name: str, 26 | in_channels: int, 27 | encoder_weights: Optional[str] = None, 28 | mixup_alpha: float = 0.5, 29 | cutmix_alpha: float = 0.5, 30 | ): 31 | super().__init__() 32 | self.cfg = cfg 33 | self.overlap = cfg.datamodule.overlap 34 | self.feature_extractor = feature_extractor 35 | self.encoder = smp.Unet( 36 | encoder_name=encoder_name, 37 | encoder_weights=encoder_weights, 38 | in_channels=in_channels, 39 | classes=1, 40 | ) 41 | self.decoder = decoder 42 | self.mixup = Mixup(mixup_alpha) 43 | self.cutmix = Cutmix(cutmix_alpha) 44 | self.loss_weight = torch.tensor(cfg.loss.loss_weight) if "loss_weight" in cfg.loss else None 45 | self.label_weight = torch.tensor(cfg.label_weight) if "label_weight" in cfg else None 46 | self.pos_weight = torch.tensor(cfg.pos_weight) if "pos_weight" in cfg else None 47 | self.loss_fn = None 48 | self.update_loss_fn() 49 | 50 | def update_loss_fn(self, sleep_decay: float = 1.0) -> None: 51 | self.label_weight[0] = self.label_weight[0] * sleep_decay 52 | if self.cfg.loss.name == "tolerance": 53 | self.loss_fn = ToleranceLoss( 54 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 55 | ) 56 | elif self.cfg.loss.name == "tolerance_mse": 57 | self.loss_fn = ToleranceMSELoss( 58 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 59 | ) 60 | elif self.cfg.loss.name == "tolerance_nonzero": 61 | self.loss_fn = ToleranceNonZeroLoss( 62 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 63 | ) 64 | elif self.cfg.loss.name == "focal": 65 | self.loss_fn = FocalLoss( 66 | alpha=self.cfg.loss.alpha, 67 | gamma=self.cfg.loss.gamma, 68 | ) 69 | elif self.cfg.loss.name == "focal_bce": 70 | self.loss_fn = FocalBCELoss( 71 | alpha=self.cfg.loss.alpha, 72 | gamma=self.cfg.loss.gamma, 73 | weight=torch.tensor(self.cfg.loss.weight), 74 | ) 75 | else: 76 | self.loss_fn = BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 77 | 78 | self.loss_fn = self.loss_fn.cuda() 79 | 80 | def forward( 81 | self, 82 | x: torch.Tensor, 83 | labels: Optional[torch.Tensor] = None, 84 | masks: Optional[torch.Tensor] = None, 85 | do_mixup: bool = False, 86 | do_cutmix: bool = False, 87 | ) -> dict[str, torch.Tensor]: 88 | """Forward pass of the model. 89 | 90 | Args: 91 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 92 | labels (Optional[torch.Tensor], optional): (batch_size, n_timesteps, n_classes) 93 | Returns: 94 | dict[str, torch.Tensor]: logits (batch_size, n_timesteps, n_classes) 95 | """ 96 | x = self.feature_extractor(x) # (batch_size, n_channels, height, n_timesteps) 97 | 98 | if do_mixup and labels is not None: 99 | x, labels = self.mixup(x, labels) 100 | if do_cutmix and labels is not None: 101 | x, labels = self.cutmix(x, labels) 102 | 103 | x = self.encoder(x).squeeze(1) # (batch_size, height, n_timesteps) 104 | logits = self.decoder(x) # (batch_size, n_timesteps, n_classes) 105 | 106 | output = {"logits": logits} 107 | if labels is not None: 108 | l = self.overlap if self.overlap > 0 else None 109 | r = -self.overlap if self.overlap > 0 else None 110 | loss = self.loss_fn(logits[:, l:r, :], labels[:, l:r, :], masks[:, l:r, :]) 111 | output["loss"] = loss 112 | 113 | return output 114 | -------------------------------------------------------------------------------- /kami/src/models/spec2Dcnn2DayV2.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | from omegaconf import DictConfig 3 | 4 | import segmentation_models_pytorch as smp 5 | import torch 6 | import torch.nn as nn 7 | 8 | from src.augmentation.cutmix import Cutmix 9 | from src.augmentation.mixup import Mixup 10 | 11 | from src.models.loss.tolerance import ToleranceLoss 12 | from src.models.loss.tolerance_mse import ToleranceMSELoss 13 | from src.models.loss.bce import BCEWithLogitsLoss 14 | from src.models.loss.tolerance_nonzero import ToleranceNonZeroLoss 15 | from src.models.loss.focal import FocalLoss 16 | from src.models.loss.focal_bce import FocalBCELoss 17 | 18 | 19 | class Spec2DCNN2DayV2(nn.Module): 20 | def __init__( 21 | self, 22 | cfg: DictConfig, 23 | feature_extractor: nn.Module, 24 | decoder: nn.Module, 25 | encoder_name: str, 26 | in_channels: int, 27 | encoder_weights: Optional[str] = None, 28 | mixup_alpha: float = 0.5, 29 | cutmix_alpha: float = 0.5, 30 | ): 31 | super().__init__() 32 | self.cfg = cfg 33 | self.feature_extractor = feature_extractor 34 | self.encoder = smp.Unet( 35 | encoder_name=encoder_name, 36 | encoder_weights=encoder_weights, 37 | in_channels=in_channels, 38 | classes=1, 39 | ) 40 | self.decoder = decoder 41 | self.mixup = Mixup(mixup_alpha) 42 | self.cutmix = Cutmix(cutmix_alpha) 43 | self.loss_weight = torch.tensor(cfg.loss.loss_weight) if "loss_weight" in cfg.loss else None 44 | self.label_weight = torch.tensor(cfg.label_weight) if "label_weight" in cfg else None 45 | self.pos_weight = torch.tensor(cfg.pos_weight) if "pos_weight" in cfg else None 46 | self.loss_fn = None 47 | self.update_loss_fn() 48 | 49 | def update_loss_fn(self, sleep_decay: float = 1.0) -> None: 50 | self.label_weight[0] = self.label_weight[0] * sleep_decay 51 | if self.cfg.loss.name == "tolerance": 52 | self.loss_fn = ToleranceLoss( 53 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 54 | ) 55 | elif self.cfg.loss.name == "tolerance_mse": 56 | self.loss_fn = ToleranceMSELoss( 57 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 58 | ) 59 | elif self.cfg.loss.name == "tolerance_nonzero": 60 | self.loss_fn = ToleranceNonZeroLoss( 61 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 62 | ) 63 | elif self.cfg.loss.name == "focal": 64 | self.loss_fn = FocalLoss( 65 | alpha=self.cfg.loss.alpha, 66 | gamma=self.cfg.loss.gamma, 67 | ) 68 | elif self.cfg.loss.name == "focal_bce": 69 | self.loss_fn = FocalBCELoss( 70 | alpha=self.cfg.loss.alpha, 71 | gamma=self.cfg.loss.gamma, 72 | weight=torch.tensor(self.cfg.loss.weight), 73 | ) 74 | else: 75 | self.loss_fn = BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 76 | 77 | self.loss_fn = self.loss_fn.cuda() 78 | 79 | def forward( 80 | self, 81 | x: torch.Tensor, 82 | labels: Optional[torch.Tensor] = None, 83 | masks: Optional[torch.Tensor] = None, 84 | do_mixup: bool = False, 85 | do_cutmix: bool = False, 86 | ) -> dict[str, torch.Tensor]: 87 | """Forward pass of the model. 88 | Args: 89 | x (torch.Tensor): (batch_size, n_channels, duration) 90 | labels (Optional[torch.Tensor], optional): (batch_size, n_timesteps, n_classes) 91 | Returns: 92 | dict[str, torch.Tensor]: logits (batch_size, n_timesteps, n_classes) 93 | """ 94 | x1, x2 = torch.split(x, x.shape[2] // 2, dim=2) 95 | 96 | # 2日分のデータを分割して並行に入力 97 | x1 = self.feature_extractor(x1) # (batch_size, n_channels, height, n_timesteps//2) 98 | x2 = self.feature_extractor(x2) # (batch_size, n_channels, height, n_timesteps//2) 99 | 100 | x1 = self.encoder(x1).squeeze(1) # (batch_size, height, n_timesteps//2) 101 | x2 = self.encoder(x2).squeeze(1) # (batch_size, height, n_timesteps//2) 102 | 103 | # 残差を結合してデコーダーに入力 104 | x = torch.cat([x1, x1 - x2], dim=1) # (batch_size, n_classes, n_timesteps//2) 105 | logits1 = self.decoder(x) # (batch_size, n_timesteps//2, n_classes) 106 | 107 | x = torch.cat([x2, x2 - x1], dim=1) 108 | logits2 = self.decoder(x) 109 | 110 | logits = torch.cat([logits1, logits2], dim=1) # (batch_size, n_timesteps, n_classes) 111 | output = {"logits": logits} 112 | if labels is not None: 113 | loss = self.loss_fn(logits, labels, masks) 114 | output["loss"] = loss 115 | 116 | return output 117 | -------------------------------------------------------------------------------- /shimacos/components/stacking/sync_batchnorm/comm.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | # File : comm.py 3 | # Author : Jiayuan Mao 4 | # Email : maojiayuan@gmail.com 5 | # Date : 27/01/2018 6 | # 7 | # This file is part of Synchronized-BatchNorm-PyTorch. 8 | # https://github.com/vacancy/Synchronized-BatchNorm-PyTorch 9 | # Distributed under MIT License. 10 | 11 | import queue 12 | import collections 13 | import threading 14 | 15 | __all__ = ['FutureResult', 'SlavePipe', 'SyncMaster'] 16 | 17 | 18 | class FutureResult(object): 19 | """A thread-safe future implementation. Used only as one-to-one pipe.""" 20 | 21 | def __init__(self): 22 | self._result = None 23 | self._lock = threading.Lock() 24 | self._cond = threading.Condition(self._lock) 25 | 26 | def put(self, result): 27 | with self._lock: 28 | assert self._result is None, 'Previous result has\'t been fetched.' 29 | self._result = result 30 | self._cond.notify() 31 | 32 | def get(self): 33 | with self._lock: 34 | if self._result is None: 35 | self._cond.wait() 36 | 37 | res = self._result 38 | self._result = None 39 | return res 40 | 41 | 42 | _MasterRegistry = collections.namedtuple('MasterRegistry', ['result']) 43 | _SlavePipeBase = collections.namedtuple('_SlavePipeBase', ['identifier', 'queue', 'result']) 44 | 45 | 46 | class SlavePipe(_SlavePipeBase): 47 | """Pipe for master-slave communication.""" 48 | 49 | def run_slave(self, msg): 50 | self.queue.put((self.identifier, msg)) 51 | ret = self.result.get() 52 | self.queue.put(True) 53 | return ret 54 | 55 | 56 | class SyncMaster(object): 57 | """An abstract `SyncMaster` object. 58 | 59 | - During the replication, as the data parallel will trigger an callback of each module, all slave devices should 60 | call `register(id)` and obtain an `SlavePipe` to communicate with the master. 61 | - During the forward pass, master device invokes `run_master`, all messages from slave devices will be collected, 62 | and passed to a registered callback. 63 | - After receiving the messages, the master device should gather the information and determine to message passed 64 | back to each slave devices. 65 | """ 66 | 67 | def __init__(self, master_callback): 68 | """ 69 | 70 | Args: 71 | master_callback: a callback to be invoked after having collected messages from slave devices. 72 | """ 73 | self._master_callback = master_callback 74 | self._queue = queue.Queue() 75 | self._registry = collections.OrderedDict() 76 | self._activated = False 77 | 78 | def __getstate__(self): 79 | return {'master_callback': self._master_callback} 80 | 81 | def __setstate__(self, state): 82 | self.__init__(state['master_callback']) 83 | 84 | def register_slave(self, identifier): 85 | """ 86 | Register an slave device. 87 | 88 | Args: 89 | identifier: an identifier, usually is the device id. 90 | 91 | Returns: a `SlavePipe` object which can be used to communicate with the master device. 92 | 93 | """ 94 | if self._activated: 95 | assert self._queue.empty(), 'Queue is not clean before next initialization.' 96 | self._activated = False 97 | self._registry.clear() 98 | future = FutureResult() 99 | self._registry[identifier] = _MasterRegistry(future) 100 | return SlavePipe(identifier, self._queue, future) 101 | 102 | def run_master(self, master_msg): 103 | """ 104 | Main entry for the master device in each forward pass. 105 | The messages were first collected from each devices (including the master device), and then 106 | an callback will be invoked to compute the message to be sent back to each devices 107 | (including the master device). 108 | 109 | Args: 110 | master_msg: the message that the master want to send to itself. This will be placed as the first 111 | message when calling `master_callback`. For detailed usage, see `_SynchronizedBatchNorm` for an example. 112 | 113 | Returns: the message to be sent back to the master device. 114 | 115 | """ 116 | self._activated = True 117 | 118 | intermediates = [(0, master_msg)] 119 | for i in range(self.nr_slaves): 120 | intermediates.append(self._queue.get()) 121 | 122 | results = self._master_callback(intermediates) 123 | assert results[0][0] == 0, 'The first result should belongs to the master.' 124 | 125 | for i, res in results: 126 | if i == 0: 127 | continue 128 | self._registry[i].result.put(res) 129 | 130 | for i in range(self.nr_slaves): 131 | assert self._queue.get() is True 132 | 133 | return results[0][1] 134 | 135 | @property 136 | def nr_slaves(self): 137 | return len(self._registry) 138 | -------------------------------------------------------------------------------- /kami/src/models/spec2DcnnAffine.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | from omegaconf import DictConfig 3 | 4 | import segmentation_models_pytorch as smp 5 | import torch 6 | import torch.nn as nn 7 | 8 | from src.augmentation.cutmix import Cutmix 9 | from src.augmentation.mixup import Mixup 10 | 11 | from src.models.loss.tolerance import ToleranceLoss 12 | from src.models.loss.tolerance_mse import ToleranceMSELoss 13 | from src.models.loss.bce import BCEWithLogitsLoss 14 | from src.models.loss.tolerance_nonzero import ToleranceNonZeroLoss 15 | from src.models.loss.focal import FocalLoss 16 | from src.models.loss.focal_bce import FocalBCELoss 17 | 18 | 19 | class Spec2DCNNAffine(nn.Module): 20 | def __init__( 21 | self, 22 | cfg: DictConfig, 23 | feature_extractor: nn.Module, 24 | decoder: nn.Module, 25 | encoder_name: str, 26 | in_channels: int, 27 | height: int, 28 | encoder_weights: Optional[str] = None, 29 | mixup_alpha: float = 0.5, 30 | cutmix_alpha: float = 0.5, 31 | ): 32 | super().__init__() 33 | self.cfg = cfg 34 | self.feature_extractor = feature_extractor 35 | aux_params = dict( 36 | pooling="avg", # one of 'avg', 'max' 37 | dropout=0.1, # dropout ratio, default is None 38 | activation=None, # activation function, default is None 39 | classes=2 * height, # define number of output labels 40 | ) 41 | 42 | self.encoder = smp.Unet( 43 | encoder_name=encoder_name, 44 | encoder_weights=encoder_weights, 45 | in_channels=in_channels, 46 | classes=1, 47 | aux_params=aux_params, 48 | ) 49 | self.decoder = decoder 50 | self.mixup = Mixup(mixup_alpha) 51 | self.cutmix = Cutmix(cutmix_alpha) 52 | self.loss_weight = torch.tensor(cfg.loss.loss_weight) if "loss_weight" in cfg.loss else None 53 | self.label_weight = torch.tensor(cfg.label_weight) if "label_weight" in cfg else None 54 | self.pos_weight = torch.tensor(cfg.pos_weight) if "pos_weight" in cfg else None 55 | self.loss_fn = None 56 | self.update_loss_fn() 57 | 58 | def update_loss_fn(self, sleep_decay: float = 1.0) -> None: 59 | self.label_weight[0] = self.label_weight[0] * sleep_decay 60 | if self.cfg.loss.name == "tolerance": 61 | self.loss_fn = ToleranceLoss( 62 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 63 | ) 64 | elif self.cfg.loss.name == "tolerance_mse": 65 | self.loss_fn = ToleranceMSELoss( 66 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 67 | ) 68 | elif self.cfg.loss.name == "tolerance_nonzero": 69 | self.loss_fn = ToleranceNonZeroLoss( 70 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 71 | ) 72 | elif self.cfg.loss.name == "focal": 73 | self.loss_fn = FocalLoss( 74 | alpha=self.cfg.loss.alpha, 75 | gamma=self.cfg.loss.gamma, 76 | ) 77 | elif self.cfg.loss.name == "focal_bce": 78 | self.loss_fn = FocalBCELoss( 79 | alpha=self.cfg.loss.alpha, 80 | gamma=self.cfg.loss.gamma, 81 | weight=torch.tensor(self.cfg.loss.weight), 82 | ) 83 | else: 84 | self.loss_fn = BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 85 | 86 | self.loss_fn = self.loss_fn.cuda() 87 | 88 | def forward( 89 | self, 90 | x: torch.Tensor, 91 | labels: Optional[torch.Tensor] = None, 92 | masks: Optional[torch.Tensor] = None, 93 | do_mixup: bool = False, 94 | do_cutmix: bool = False, 95 | ) -> dict[str, torch.Tensor]: 96 | """Forward pass of the model. 97 | 98 | Args: 99 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 100 | labels (Optional[torch.Tensor], optional): (batch_size, n_timesteps, n_classes) 101 | Returns: 102 | dict[str, torch.Tensor]: logits (batch_size, n_timesteps, n_classes) 103 | """ 104 | x = self.feature_extractor(x) # (batch_size, n_channels, height, n_timesteps) 105 | 106 | if do_mixup and labels is not None: 107 | x, labels = self.mixup(x, labels) 108 | if do_cutmix and labels is not None: 109 | x, labels = self.cutmix(x, labels) 110 | 111 | x, mid = self.encoder(x) # (batch_size, 1, height, n_timesteps), (batch_size, 2 * height) 112 | a, b = torch.split(mid, mid.shape[1] // 2, dim=1) 113 | x = x.squeeze(1) # (batch_size, height, n_timesteps) 114 | x = a.unsqueeze(1) * x + b.unsqueeze(1) # (batch_size, height, n_timesteps) 115 | logits = self.decoder(x) # (batch_size, n_classes, n_timesteps) 116 | 117 | output = {"logits": logits} 118 | if labels is not None: 119 | loss = self.loss_fn(logits, labels, masks) 120 | output["loss"] = loss 121 | 122 | return output 123 | -------------------------------------------------------------------------------- /requirements.lock: -------------------------------------------------------------------------------- 1 | # generated by rye 2 | # use `rye lock` or `rye sync` to update this lockfile 3 | # 4 | # last locked with the following flags: 5 | # pre: false 6 | # features: [] 7 | # all-features: false 8 | 9 | -e file:. 10 | aiohttp==3.8.5 11 | aiosignal==1.3.1 12 | aliyun-python-sdk-core==2.13.36 13 | aliyun-python-sdk-kms==2.16.1 14 | antlr4-python3-runtime==4.9.3 15 | anyio==3.7.1 16 | appdirs==1.4.4 17 | argon2-cffi==23.1.0 18 | argon2-cffi-bindings==21.2.0 19 | arrow==1.2.3 20 | asttokens==2.2.1 21 | async-lru==2.0.4 22 | async-timeout==4.0.3 23 | attrs==23.1.0 24 | babel==2.12.1 25 | backcall==0.2.0 26 | beautifulsoup4==4.12.2 27 | bleach==6.0.0 28 | cachetools==5.3.1 29 | catboost==1.2 30 | certifi==2023.7.22 31 | cffi==1.15.1 32 | charset-normalizer==3.2.0 33 | click==8.1.6 34 | cmake==3.27.2 35 | colorama==0.4.6 36 | colorlog==6.7.0 37 | comm==0.1.4 38 | contourpy==1.1.0 39 | crcmod==1.7 40 | cryptography==41.0.3 41 | cycler==0.11.0 42 | debugpy==1.6.7.post1 43 | decorator==5.1.1 44 | deepspeed==0.10.0 45 | defusedxml==0.7.1 46 | docker-pycreds==0.4.0 47 | efficientnet-pytorch==0.7.1 48 | exceptiongroup==1.1.3 49 | executing==1.2.0 50 | fairscale==0.4.13 51 | fastjsonschema==2.18.0 52 | filelock==3.12.2 53 | fonttools==4.42.0 54 | fqdn==1.5.1 55 | frozenlist==1.4.0 56 | fsspec==2023.6.0 57 | gitdb==4.0.10 58 | gitpython==3.1.32 59 | google-api-core==2.11.1 60 | google-auth==2.22.0 61 | google-cloud-core==2.3.3 62 | google-cloud-storage==2.10.0 63 | google-crc32c==1.5.0 64 | google-resumable-media==2.5.0 65 | googleapis-common-protos==1.60.0 66 | graphviz==0.20.1 67 | hjson==3.1.0 68 | huggingface-hub==0.16.4 69 | hydra-colorlog==1.2.0 70 | hydra-core==1.3.2 71 | idna==3.4 72 | invoke==2.2.0 73 | ipykernel==6.25.1 74 | ipython==8.14.0 75 | ipython-genutils==0.2.0 76 | ipywidgets==8.1.0 77 | isoduration==20.11.0 78 | jedi==0.19.0 79 | jinja2==3.1.2 80 | jmespath==0.10.0 81 | joblib==1.3.2 82 | json5==0.9.14 83 | jsonpointer==2.4 84 | jsonschema==4.19.0 85 | jsonschema-specifications==2023.7.1 86 | jupyter==1.0.0 87 | jupyter-client==8.3.0 88 | jupyter-console==6.6.3 89 | jupyter-core==5.3.1 90 | jupyter-events==0.7.0 91 | jupyter-lsp==2.2.0 92 | jupyter-server==2.7.1 93 | jupyter-server-terminals==0.4.4 94 | jupyterlab==4.0.5 95 | jupyterlab-pygments==0.2.2 96 | jupyterlab-server==2.24.0 97 | jupyterlab-widgets==3.0.8 98 | kiwisolver==1.4.4 99 | lightgbm==4.0.0 100 | lightning-utilities==0.9.0 101 | lit==16.0.6 102 | llvmlite==0.41.1 103 | loguru==0.7.2 104 | markdown==3.4.4 105 | markdown-it-py==3.0.0 106 | markupsafe==2.1.3 107 | matplotlib==3.7.2 108 | matplotlib-inline==0.1.6 109 | matplotlib-venn==0.11.9 110 | mdurl==0.1.2 111 | mistune==3.0.1 112 | model-index==0.1.11 113 | mpmath==1.3.0 114 | multidict==6.0.4 115 | munch==4.0.0 116 | nbclient==0.8.0 117 | nbconvert==7.7.3 118 | nbformat==5.9.2 119 | nest-asyncio==1.5.7 120 | networkx==3.1 121 | ninja==1.11.1 122 | notebook==7.0.2 123 | notebook-shim==0.2.3 124 | numba==0.58.1 125 | numpy==1.25.2 126 | omegaconf==2.3.0 127 | opendatalab==0.0.10 128 | openmim==0.3.9 129 | openxlab==0.0.17 130 | ordered-set==4.1.0 131 | oss2==2.17.0 132 | overrides==7.4.0 133 | packaging==23.1 134 | pandas==2.0.3 135 | pandocfilters==1.5.0 136 | parso==0.8.3 137 | pathlib==1.0.1 138 | pathtools==0.1.2 139 | pexpect==4.8.0 140 | pickleshare==0.7.5 141 | pillow==9.5.0 142 | platformdirs==3.10.0 143 | plotly==5.16.0 144 | polars==0.19.13 145 | pretrainedmodels==0.7.4 146 | prometheus-client==0.17.1 147 | prompt-toolkit==3.0.39 148 | protobuf==4.24.0 149 | psutil==5.9.5 150 | ptyprocess==0.7.0 151 | pure-eval==0.2.2 152 | py-cpuinfo==9.0.0 153 | pyarrow==14.0.1 154 | pyasn1==0.5.0 155 | pyasn1-modules==0.3.0 156 | pycparser==2.21 157 | pycryptodome==3.18.0 158 | pydantic==1.10.12 159 | pygments==2.16.1 160 | pyparsing==3.0.9 161 | python-dateutil==2.8.2 162 | python-json-logger==2.0.7 163 | pytorch-lightning==2.0.6 164 | pytz==2023.3 165 | pyyaml==6.0.1 166 | pyzmq==25.1.1 167 | qtconsole==5.4.3 168 | qtpy==2.3.1 169 | referencing==0.30.2 170 | regex==2023.8.8 171 | requests==2.28.2 172 | rfc3339-validator==0.1.4 173 | rfc3986-validator==0.1.1 174 | rich==13.4.2 175 | rpds-py==0.9.2 176 | rsa==4.9 177 | safetensors==0.3.2 178 | scikit-learn==1.3.0 179 | scipy==1.11.4 180 | seaborn==0.12.2 181 | segmentation-models-pytorch==0.3.3 182 | send2trash==1.8.2 183 | sentry-sdk==1.29.2 184 | seqeval==1.2.2 185 | setproctitle==1.3.2 186 | six==1.16.0 187 | smmap==5.0.0 188 | sniffio==1.3.0 189 | soupsieve==2.4.1 190 | stack-data==0.6.2 191 | sympy==1.12 192 | tabulate==0.9.0 193 | tenacity==8.2.3 194 | terminado==0.17.1 195 | threadpoolctl==3.2.0 196 | timm==0.9.2 197 | tinycss2==1.2.1 198 | tokenizers==0.13.3 199 | tomli==2.0.1 200 | torch==2.0.1+cu118 201 | torchaudio==2.0.2+cu118 202 | torchmetrics==1.0.3 203 | torchvision==0.15.2+cu118 204 | tornado==6.3.3 205 | tqdm==4.65.2 206 | traitlets==5.9.0 207 | transformers==4.31.0 208 | triton==2.0.0 209 | typing-extensions==4.7.1 210 | tzdata==2023.3 211 | uri-template==1.3.0 212 | urllib3==1.26.16 213 | wandb==0.15.8 214 | wcwidth==0.2.6 215 | webcolors==1.13 216 | webencodings==0.5.1 217 | websocket-client==1.6.1 218 | widgetsnbextension==4.0.8 219 | xgboost==1.7.6 220 | yarl==1.9.2 221 | # The following packages are considered to be unsafe in a requirements file: 222 | pip==23.2.1 223 | setuptools==60.2.0 224 | -------------------------------------------------------------------------------- /kami/src/models/spec2Dcnn2Day.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | from omegaconf import DictConfig 3 | 4 | import segmentation_models_pytorch as smp 5 | import torch 6 | import torch.nn as nn 7 | 8 | from src.augmentation.cutmix import Cutmix 9 | from src.augmentation.mixup import Mixup 10 | 11 | from src.models.loss.tolerance import ToleranceLoss 12 | from src.models.loss.tolerance_mse import ToleranceMSELoss 13 | from src.models.loss.bce import BCEWithLogitsLoss 14 | from src.models.loss.tolerance_nonzero import ToleranceNonZeroLoss 15 | from src.models.loss.focal import FocalLoss 16 | from src.models.loss.focal_bce import FocalBCELoss 17 | 18 | 19 | class Spec2DCNN2Day(nn.Module): 20 | def __init__( 21 | self, 22 | cfg: DictConfig, 23 | feature_extractor: nn.Module, 24 | decoder: nn.Module, 25 | encoder_name: str, 26 | in_channels: int, 27 | encoder_weights: Optional[str] = None, 28 | mixup_alpha: float = 0.5, 29 | cutmix_alpha: float = 0.5, 30 | ): 31 | super().__init__() 32 | self.cfg = cfg 33 | self.feature_extractor = feature_extractor 34 | self.encoder = smp.Unet( 35 | encoder_name=encoder_name, 36 | encoder_weights=encoder_weights, 37 | in_channels=in_channels, 38 | classes=1, 39 | ) 40 | self.decoder = decoder 41 | self.mixup = Mixup(mixup_alpha) 42 | self.cutmix = Cutmix(cutmix_alpha) 43 | self.loss_weight = torch.tensor(cfg.loss.loss_weight) if "loss_weight" in cfg.loss else None 44 | self.label_weight = torch.tensor(cfg.label_weight) if "label_weight" in cfg else None 45 | self.pos_weight = torch.tensor(cfg.pos_weight) if "pos_weight" in cfg else None 46 | self.loss_fn = None 47 | self.update_loss_fn() 48 | 49 | def update_loss_fn(self, sleep_decay: float = 1.0) -> None: 50 | self.label_weight[0] = self.label_weight[0] * sleep_decay 51 | if self.cfg.loss.name == "tolerance": 52 | self.loss_fn = ToleranceLoss( 53 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 54 | ) 55 | elif self.cfg.loss.name == "tolerance_mse": 56 | self.loss_fn = ToleranceMSELoss( 57 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 58 | ) 59 | elif self.cfg.loss.name == "tolerance_nonzero": 60 | self.loss_fn = ToleranceNonZeroLoss( 61 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 62 | ) 63 | elif self.cfg.loss.name == "focal": 64 | self.loss_fn = FocalLoss( 65 | alpha=self.cfg.loss.alpha, 66 | gamma=self.cfg.loss.gamma, 67 | ) 68 | elif self.cfg.loss.name == "focal_bce": 69 | self.loss_fn = FocalBCELoss( 70 | alpha=self.cfg.loss.alpha, 71 | gamma=self.cfg.loss.gamma, 72 | weight=torch.tensor(self.cfg.loss.weight), 73 | ) 74 | else: 75 | self.loss_fn = BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 76 | 77 | self.loss_fn = self.loss_fn.cuda() 78 | 79 | def forward( 80 | self, 81 | x: torch.Tensor, 82 | labels: Optional[torch.Tensor] = None, 83 | masks: Optional[torch.Tensor] = None, 84 | do_mixup: bool = False, 85 | do_cutmix: bool = False, 86 | ) -> dict[str, torch.Tensor]: 87 | """Forward pass of the model. 88 | 89 | Args: 90 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 91 | labels (Optional[torch.Tensor], optional): (batch_size, n_timesteps, n_classes) 92 | Returns: 93 | dict[str, torch.Tensor]: logits (batch_size, n_timesteps, n_classes) 94 | """ 95 | x1, x2 = torch.split(x, x.shape[2] // 2, dim=2) 96 | 97 | # 2日分のデータを分割して並行に入力 98 | x1 = self.feature_extractor(x1) # (batch_size, n_channels, height, n_timesteps) 99 | x2 = self.feature_extractor(x2) # (batch_size, n_channels, height, n_timesteps) 100 | 101 | x1 = self.encoder(x1).squeeze(1) # (batch_size, height, n_timesteps) 102 | x2 = self.encoder(x2).squeeze(1) # (batch_size, height, n_timesteps) 103 | 104 | # 2日分のデータを結合 105 | x = torch.cat([x1, x2], dim=1) # (batch_size, 2 * n_classes, n_timesteps) 106 | logits1 = self.decoder(x) # (batch_size, n_timesteps, 2 * n_classes) 107 | l1, l2 = torch.split(logits1, logits1.shape[2] // 2, dim=2) 108 | logits1 = torch.concat([l1, l2], dim=1) # (batch_size, 2*n_timesteps, n_classes) 109 | 110 | x = torch.cat([x2, x1], dim=1) 111 | logits2 = self.decoder(x) 112 | l2, l1 = torch.split(logits2, logits2.shape[2] // 2, dim=2) 113 | logits2 = torch.concat([l1, l2], dim=1) 114 | 115 | logits = (logits1 + logits2) / 2 116 | output = {"logits": logits} 117 | if labels is not None: 118 | loss1 = self.loss_fn(logits1, labels, masks) 119 | loss2 = self.loss_fn(logits2, labels, masks) 120 | output["loss"] = (loss1 + loss2) / 2 121 | 122 | return output 123 | -------------------------------------------------------------------------------- /kami/src/models/spec2DcnnMinMax.py: -------------------------------------------------------------------------------- 1 | from typing import Optional 2 | from omegaconf import DictConfig 3 | 4 | import segmentation_models_pytorch as smp 5 | import torch 6 | import torch.nn as nn 7 | 8 | from src.augmentation.cutmix import Cutmix 9 | from src.augmentation.mixup import Mixup 10 | 11 | from src.models.loss.tolerance import ToleranceLoss 12 | from src.models.loss.tolerance_mse import ToleranceMSELoss 13 | from src.models.loss.bce import BCEWithLogitsLoss 14 | from src.models.loss.tolerance_nonzero import ToleranceNonZeroLoss 15 | from src.models.loss.focal import FocalLoss 16 | from src.models.loss.focal_bce import FocalBCELoss 17 | 18 | 19 | class Spec2DCNNMinMax(nn.Module): 20 | def __init__( 21 | self, 22 | cfg: DictConfig, 23 | feature_extractor: nn.Module, 24 | decoder: nn.Module, 25 | encoder_name: str, 26 | in_channels: int, 27 | height: int, 28 | encoder_weights: Optional[str] = None, 29 | mixup_alpha: float = 0.5, 30 | cutmix_alpha: float = 0.5, 31 | ): 32 | super().__init__() 33 | self.cfg = cfg 34 | self.feature_extractor = feature_extractor 35 | aux_params = dict( 36 | pooling="avg", # one of 'avg', 'max' 37 | dropout=0.1, # dropout ratio, default is None 38 | activation=None, # activation function, default is None 39 | classes=2 * height, # define number of output labels 40 | ) 41 | 42 | self.encoder = smp.Unet( 43 | encoder_name=encoder_name, 44 | encoder_weights=encoder_weights, 45 | in_channels=in_channels, 46 | classes=1, 47 | aux_params=aux_params, 48 | ) 49 | self.decoder = decoder 50 | self.mixup = Mixup(mixup_alpha) 51 | self.cutmix = Cutmix(cutmix_alpha) 52 | self.loss_weight = torch.tensor(cfg.loss.loss_weight) if "loss_weight" in cfg.loss else None 53 | self.label_weight = torch.tensor(cfg.label_weight) if "label_weight" in cfg else None 54 | self.pos_weight = torch.tensor(cfg.pos_weight) if "pos_weight" in cfg else None 55 | self.loss_fn = None 56 | self.update_loss_fn() 57 | 58 | def update_loss_fn(self, sleep_decay: float = 1.0) -> None: 59 | self.label_weight[0] = self.label_weight[0] * sleep_decay 60 | if self.cfg.loss.name == "tolerance": 61 | self.loss_fn = ToleranceLoss( 62 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 63 | ) 64 | elif self.cfg.loss.name == "tolerance_mse": 65 | self.loss_fn = ToleranceMSELoss( 66 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 67 | ) 68 | elif self.cfg.loss.name == "tolerance_nonzero": 69 | self.loss_fn = ToleranceNonZeroLoss( 70 | loss_weight=self.loss_weight, label_weight=self.label_weight, pos_weight=self.pos_weight 71 | ) 72 | elif self.cfg.loss.name == "focal": 73 | self.loss_fn = FocalLoss( 74 | alpha=self.cfg.loss.alpha, 75 | gamma=self.cfg.loss.gamma, 76 | ) 77 | elif self.cfg.loss.name == "focal_bce": 78 | self.loss_fn = FocalBCELoss( 79 | alpha=self.cfg.loss.alpha, 80 | gamma=self.cfg.loss.gamma, 81 | weight=torch.tensor(self.cfg.loss.weight), 82 | ) 83 | else: 84 | self.loss_fn = BCEWithLogitsLoss(weight=self.label_weight, pos_weight=self.pos_weight) 85 | 86 | self.loss_fn = self.loss_fn.cuda() 87 | 88 | def forward( 89 | self, 90 | x: torch.Tensor, 91 | labels: Optional[torch.Tensor] = None, 92 | masks: Optional[torch.Tensor] = None, 93 | do_mixup: bool = False, 94 | do_cutmix: bool = False, 95 | ) -> dict[str, torch.Tensor]: 96 | """Forward pass of the model. 97 | 98 | Args: 99 | x (torch.Tensor): (batch_size, n_channels, n_timesteps) 100 | labels (Optional[torch.Tensor], optional): (batch_size, n_timesteps, n_classes) 101 | Returns: 102 | dict[str, torch.Tensor]: logits (batch_size, n_timesteps, n_classes) 103 | """ 104 | x = self.feature_extractor(x) # (batch_size, n_channels, height, n_timesteps) 105 | 106 | if do_mixup and labels is not None: 107 | x, labels = self.mixup(x, labels) 108 | if do_cutmix and labels is not None: 109 | x, labels = self.cutmix(x, labels) 110 | 111 | x, mid = self.encoder(x) # (batch_size, 1, height, n_timesteps), (batch_size, 2 * height) 112 | x_min, x_max = torch.split(mid, mid.shape[1] // 2, dim=1) 113 | x = x.squeeze(1) # (batch_size, height, n_timesteps) 114 | x = torch.cat( 115 | [x, x - x_min.unsqueeze(2), x_max.unsqueeze(2) - x], dim=1 116 | ) # (batch_size, 3 * height, n_timesteps) 117 | logits = self.decoder(x) # (batch_size, n_classes, n_timesteps) 118 | 119 | output = {"logits": logits} 120 | if labels is not None: 121 | loss = self.loss_fn(logits, labels, masks) 122 | output["loss"] = loss 123 | 124 | return output 125 | -------------------------------------------------------------------------------- /requirements-dev.lock: -------------------------------------------------------------------------------- 1 | # generated by rye 2 | # use `rye lock` or `rye sync` to update this lockfile 3 | # 4 | # last locked with the following flags: 5 | # pre: false 6 | # features: [] 7 | # all-features: false 8 | 9 | -e file:. 10 | aiohttp==3.8.5 11 | aiosignal==1.3.1 12 | aliyun-python-sdk-core==2.13.36 13 | aliyun-python-sdk-kms==2.16.1 14 | antlr4-python3-runtime==4.9.3 15 | anyio==3.7.1 16 | appdirs==1.4.4 17 | argon2-cffi==23.1.0 18 | argon2-cffi-bindings==21.2.0 19 | arrow==1.2.3 20 | asttokens==2.2.1 21 | async-lru==2.0.4 22 | async-timeout==4.0.3 23 | attrs==23.1.0 24 | babel==2.12.1 25 | backcall==0.2.0 26 | beautifulsoup4==4.12.2 27 | black==23.7.0 28 | bleach==6.0.0 29 | cachetools==5.3.1 30 | catboost==1.2 31 | certifi==2023.7.22 32 | cffi==1.15.1 33 | charset-normalizer==3.2.0 34 | click==8.1.6 35 | cmake==3.27.2 36 | colorama==0.4.6 37 | colorlog==6.7.0 38 | comm==0.1.4 39 | contourpy==1.1.0 40 | crcmod==1.7 41 | cryptography==41.0.3 42 | cycler==0.11.0 43 | debugpy==1.6.7.post1 44 | decorator==5.1.1 45 | deepspeed==0.10.0 46 | defusedxml==0.7.1 47 | docker-pycreds==0.4.0 48 | efficientnet-pytorch==0.7.1 49 | exceptiongroup==1.1.3 50 | executing==1.2.0 51 | fairscale==0.4.13 52 | fastjsonschema==2.18.0 53 | filelock==3.12.2 54 | flake8==6.1.0 55 | fonttools==4.42.0 56 | fqdn==1.5.1 57 | frozenlist==1.4.0 58 | fsspec==2023.6.0 59 | gitdb==4.0.10 60 | gitpython==3.1.32 61 | google-api-core==2.11.1 62 | google-auth==2.22.0 63 | google-cloud-core==2.3.3 64 | google-cloud-storage==2.10.0 65 | google-crc32c==1.5.0 66 | google-resumable-media==2.5.0 67 | googleapis-common-protos==1.60.0 68 | graphviz==0.20.1 69 | hjson==3.1.0 70 | huggingface-hub==0.16.4 71 | hydra-colorlog==1.2.0 72 | hydra-core==1.3.2 73 | idna==3.4 74 | invoke==2.2.0 75 | ipykernel==6.25.1 76 | ipython==8.14.0 77 | ipython-genutils==0.2.0 78 | ipywidgets==8.1.0 79 | isoduration==20.11.0 80 | jedi==0.19.0 81 | jinja2==3.1.2 82 | jmespath==0.10.0 83 | joblib==1.3.2 84 | json5==0.9.14 85 | jsonpointer==2.4 86 | jsonschema==4.19.0 87 | jsonschema-specifications==2023.7.1 88 | jupyter==1.0.0 89 | jupyter-client==8.3.0 90 | jupyter-console==6.6.3 91 | jupyter-core==5.3.1 92 | jupyter-events==0.7.0 93 | jupyter-lsp==2.2.0 94 | jupyter-server==2.7.1 95 | jupyter-server-terminals==0.4.4 96 | jupyterlab==4.0.5 97 | jupyterlab-pygments==0.2.2 98 | jupyterlab-server==2.24.0 99 | jupyterlab-widgets==3.0.8 100 | kaggle==1.5.16 101 | kiwisolver==1.4.4 102 | lightgbm==4.0.0 103 | lightning-utilities==0.9.0 104 | lit==16.0.6 105 | llvmlite==0.41.1 106 | loguru==0.7.2 107 | markdown==3.4.4 108 | markdown-it-py==3.0.0 109 | markupsafe==2.1.3 110 | matplotlib==3.7.2 111 | matplotlib-inline==0.1.6 112 | matplotlib-venn==0.11.9 113 | mccabe==0.7.0 114 | mdurl==0.1.2 115 | mistune==3.0.1 116 | model-index==0.1.11 117 | mpmath==1.3.0 118 | multidict==6.0.4 119 | munch==4.0.0 120 | mypy==1.5.0 121 | mypy-extensions==1.0.0 122 | nbclient==0.8.0 123 | nbconvert==7.7.3 124 | nbformat==5.9.2 125 | nest-asyncio==1.5.7 126 | networkx==3.1 127 | ninja==1.11.1 128 | notebook==7.0.2 129 | notebook-shim==0.2.3 130 | numba==0.58.1 131 | numpy==1.25.2 132 | omegaconf==2.3.0 133 | opendatalab==0.0.10 134 | openmim==0.3.9 135 | openxlab==0.0.17 136 | ordered-set==4.1.0 137 | oss2==2.17.0 138 | overrides==7.4.0 139 | packaging==23.1 140 | pandas==2.0.3 141 | pandocfilters==1.5.0 142 | parso==0.8.3 143 | pathlib==1.0.1 144 | pathspec==0.11.2 145 | pathtools==0.1.2 146 | pexpect==4.8.0 147 | pickleshare==0.7.5 148 | pillow==9.5.0 149 | platformdirs==3.10.0 150 | plotly==5.16.0 151 | polars==0.19.13 152 | pretrainedmodels==0.7.4 153 | prometheus-client==0.17.1 154 | prompt-toolkit==3.0.39 155 | protobuf==4.24.0 156 | psutil==5.9.5 157 | ptyprocess==0.7.0 158 | pure-eval==0.2.2 159 | py-cpuinfo==9.0.0 160 | pyarrow==14.0.1 161 | pyasn1==0.5.0 162 | pyasn1-modules==0.3.0 163 | pycodestyle==2.11.0 164 | pycparser==2.21 165 | pycryptodome==3.18.0 166 | pydantic==1.10.12 167 | pyflakes==3.1.0 168 | pygments==2.16.1 169 | pyparsing==3.0.9 170 | python-dateutil==2.8.2 171 | python-json-logger==2.0.7 172 | python-slugify==8.0.1 173 | pytorch-lightning==2.0.6 174 | pytz==2023.3 175 | pyyaml==6.0.1 176 | pyzmq==25.1.1 177 | qtconsole==5.4.3 178 | qtpy==2.3.1 179 | referencing==0.30.2 180 | regex==2023.8.8 181 | requests==2.28.2 182 | rfc3339-validator==0.1.4 183 | rfc3986-validator==0.1.1 184 | rich==13.4.2 185 | rpds-py==0.9.2 186 | rsa==4.9 187 | ruff==0.0.284 188 | safetensors==0.3.2 189 | scikit-learn==1.3.0 190 | scipy==1.11.4 191 | seaborn==0.12.2 192 | segmentation-models-pytorch==0.3.3 193 | send2trash==1.8.2 194 | sentry-sdk==1.29.2 195 | seqeval==1.2.2 196 | setproctitle==1.3.2 197 | six==1.16.0 198 | smmap==5.0.0 199 | sniffio==1.3.0 200 | soupsieve==2.4.1 201 | stack-data==0.6.2 202 | sympy==1.12 203 | tabulate==0.9.0 204 | tenacity==8.2.3 205 | terminado==0.17.1 206 | text-unidecode==1.3 207 | threadpoolctl==3.2.0 208 | timm==0.9.2 209 | tinycss2==1.2.1 210 | tokenizers==0.13.3 211 | tomli==2.0.1 212 | torch==2.0.1+cu118 213 | torchaudio==2.0.2+cu118 214 | torchmetrics==1.0.3 215 | torchvision==0.15.2+cu118 216 | tornado==6.3.3 217 | tqdm==4.65.2 218 | traitlets==5.9.0 219 | transformers==4.31.0 220 | triton==2.0.0 221 | typing-extensions==4.7.1 222 | tzdata==2023.3 223 | uri-template==1.3.0 224 | urllib3==1.26.16 225 | wandb==0.15.8 226 | wcwidth==0.2.6 227 | webcolors==1.13 228 | webencodings==0.5.1 229 | websocket-client==1.6.1 230 | widgetsnbextension==4.0.8 231 | xgboost==1.7.6 232 | yarl==1.9.2 233 | # The following packages are considered to be unsafe in a requirements file: 234 | pip==23.2.1 235 | setuptools==60.2.0 236 | -------------------------------------------------------------------------------- /input/folds/stratify_fold_0.yaml: -------------------------------------------------------------------------------- 1 | train_series_ids: 2 | - 72bbd1ac3edf 3 | - d93b0c7de16b 4 | - 76237b9406d5 5 | - 2b0a1fa8eba8 6 | - 1087d7b0ff2e 7 | - 8b159a98f485 8 | - bfe41e96d12f 9 | - 8a22387617c3 10 | - 77ca4db83644 11 | - 7476c0bd18d2 12 | - 1d4569cbac0f 13 | - a9e5f5314bcb 14 | - f981a0805fd0 15 | - de6fedfb6139 16 | - 7df249527c63 17 | - d2d6b9af0553 18 | - 694faf956ebf 19 | - b364205aba43 20 | - fcca183903b7 21 | - 3d53bfea61d6 22 | - 55a47ff9dc8a 23 | - 35826366dfc7 24 | - ca732a3c37f7 25 | - fe90110788d2 26 | - d2fef7e4defd 27 | - 72ba4a8afff4 28 | - 4743bdde25df 29 | - ece2561f07e9 30 | - 90eac42a9ec9 31 | - c107b5789660 32 | - 0dee4fda51c3 33 | - 5acc9d63b5fd 34 | - 99b829cbad2d 35 | - db5e0ee1c0ab 36 | - 4b45c36f8f5a 37 | - e867b5133665 38 | - f6d2cc003183 39 | - 752900afe3a6 40 | - eef041dd50aa 41 | - 349c5562ee2c 42 | - 5ffd5e1e81ac 43 | - 29c75c018220 44 | - e2a849d283c0 45 | - dfc3ccebfdc9 46 | - d9e887091a5c 47 | - 062dbd4c95e6 48 | - e0d7b0dcf9f3 49 | - 3be2f86c3e45 50 | - c68260cc9e8f 51 | - ebb6fae8ed43 52 | - cf13ed7e457a 53 | - 03d92c9f6f8a 54 | - 6d6b9d22d48a 55 | - ca730dbf521d 56 | - 7504165f497d 57 | - b750c8c1556c 58 | - 038441c925bb 59 | - 44a41bba1ee7 60 | - a88088855de5 61 | - 6ca4f4fca6a2 62 | - 40dce6018935 63 | - 8f6f15b9f598 64 | - 0cfc06c129cc 65 | - 29d3469bd15d 66 | - 1e6717d93c1d 67 | - c289c8a823e0 68 | - 51b23d177971 69 | - dff367373725 70 | - a9a2f7fac455 71 | - 67f5fc60e494 72 | - 3318a0e3ed6f 73 | - dacc6d652e35 74 | - 207eded97727 75 | - f564985ab692 76 | - 55b7f5c99930 77 | - fa149c3c4bde 78 | - 9ddd40f2cb36 79 | - 2fbbee1a38e3 80 | - bf00506437aa 81 | - 12d01911d509 82 | - c38707ef76df 83 | - 33ceeba8918a 84 | - bb5612895813 85 | - 4ac356361be9 86 | - e0686434d029 87 | - 6ee4ade1f2bd 88 | - ccdee561ee5d 89 | - 72d2234e84e4 90 | - b737f8c78ec5 91 | - b7188813d58a 92 | - ebd76e93ec7d 93 | - a2b0a64ec9cf 94 | - 60e51cad2ffb 95 | - d3dddd3c0e00 96 | - bccf2f2819f8 97 | - ce9164297046 98 | - 1b92be89db4c 99 | - c7d693f24684 100 | - d5e47b94477e 101 | - 405df1b41f9f 102 | - 939932f1822d 103 | - 0a96f4993bd7 104 | - c8053490cec2 105 | - e1f5abb82285 106 | - 8b8b9e29171c 107 | - eec197a4bdca 108 | - 10469f6765bf 109 | - 3664fe9233f9 110 | - 1716cd4163b2 111 | - e4500e7e19e1 112 | - e30cb792a2bc 113 | - 60d31b0bec3b 114 | - d150801f3145 115 | - 8877a6586606 116 | - 844f54dcab89 117 | - 0f572d690310 118 | - 9277be28a1cf 119 | - 27f09a6a858f 120 | - e1f2a4f991cb 121 | - 655f19eabf1e 122 | - 7fd4284b7ee8 123 | - 612aa8ba44e2 124 | - 0cd1e3d0ed95 125 | - aa81faa78747 126 | - 25e2b3dd9c3b 127 | - 5aad18e7ce64 128 | - 0d0ad1e77851 129 | - 5c088d7e916c 130 | - d043c0ca71cd 131 | - ba8083a2c3b8 132 | - e6ddbaaf0639 133 | - 137b99e936ab 134 | - d8de352c2657 135 | - 653622ac8363 136 | - fbf33b1a2c10 137 | - fb223ed2278c 138 | - dc80ca623d71 139 | - 5e816f11f5c3 140 | - 5f94bb3e1bed 141 | - 483d6545417f 142 | - 3a9a9dc2cbd9 143 | - ad425f3ee76d 144 | - a81f4472c637 145 | - 3df0da2e5966 146 | - a261bc4b7470 147 | - c908a0ad3e31 148 | - 416354edd92a 149 | - 78569a801a38 150 | - 062cae666e2a 151 | - b84960841a75 152 | - d25e479ecbb7 153 | - 10f8bc1f7b07 154 | - 154fe824ed87 155 | - 99237ce045e4 156 | - 6a4cd123bd69 157 | - 1762ab70ec76 158 | - 559ffb7c166a 159 | - 599ca4ed791b 160 | - 971207c6a525 161 | - f0482490923c 162 | - 3be1545083b7 163 | - a167532acca2 164 | - 2654a87be968 165 | - f88e18cb4100 166 | - 9fbdeffbe2ba 167 | - 16fe2798ed0f 168 | - 4ab54be1a403 169 | - c6788e579967 170 | - 137771d19ca2 171 | - 91127c2b0e60 172 | - d5be621fd9aa 173 | - efbfc4526d58 174 | - 2cd2340ca14d 175 | - ee4e0e3afd3d 176 | - 5c55a5e717d6 177 | - 449766346eb1 178 | - f2c2436cf7b7 179 | - af91d9a50547 180 | - f56824b503a0 181 | - 31011ade7c0a 182 | - 6bf95a3cf91c 183 | - 188d4b7cd28b 184 | - e586cbfa7762 185 | - 4a31811f3558 186 | - 927dd0c35dfd 187 | - c5365a55ebb7 188 | - 08db4255286f 189 | - 9b9cd7b7af8c 190 | - a681f9b04b21 191 | - 361366da569e 192 | - 0ec9fc461819 193 | - 44d8c02b369e 194 | - 83fa182bec3a 195 | - 804594bb1f06 196 | - 8becc76ea607 197 | - 519ae2d858b0 198 | - ea0770830757 199 | - e8d0a37c3eba 200 | - 2b8d87addea9 201 | - f7eb179216c2 202 | - 8e32047cbc1f 203 | - c75b4b207bea 204 | - 51fdcc8d9fe7 205 | - 1c7c0bad1263 206 | - 9aed9ee12ae2 207 | - df33ae359fb5 208 | - 87a6cbb7c4ed 209 | - bdfce9ce62b9 210 | - 18a0ca03431d 211 | - 703b5efa9bc1 212 | - 0ce74d6d2106 213 | - 8fb18e36697d 214 | - 148471991ffb 215 | - e69aff66e0cb 216 | - c3072a759efb 217 | - c5d08fc3e040 218 | - c7b1283bb7eb 219 | - 0f9e60a8e56d 220 | - 390b487231ce 221 | - e11b9d69f856 222 | - 89c7daa72eee 223 | valid_series_ids: 224 | - bfa54bd26187 225 | - d515236bdeec 226 | - 04f547b8017d 227 | - 51c49c540b4e 228 | - 8898e6db816d 229 | - 18b61dd5aae8 230 | - 1319a1935f48 231 | - 1955d568d987 232 | - e34b496b84ce 233 | - 0ef7d94fde99 234 | - 280e08693c6d 235 | - 2e9ced2c7976 236 | - 292a75c0b94e 237 | - f8a8da8bdd00 238 | - 3c336d6ba566 239 | - 89bd631d1769 240 | - 3452b878e596 241 | - b4b75225b224 242 | - 9ee455e4770d 243 | - a4e48102f402 244 | - 3aceb17ef7bd 245 | - e2b60820c325 246 | - b1831c4979da 247 | - 2f7504d0f426 248 | - c7b2155a4a47 249 | - 785c9ca4eff7 250 | - 854206f602d0 251 | - 13b4d6a01d27 252 | - 05e1944c3818 253 | - aed3850f65f0 254 | - 0402a003dae9 255 | - d0f613c700f7 256 | - b7fc34995d0f 257 | - 601559e1777d 258 | - 702bb5387b1e 259 | - cca14d1966c1 260 | - a596ad0b82aa 261 | - 8a306e0890c0 262 | - 91cb6c98201f 263 | - 9a340507e36a 264 | - cfeb11428dd7 265 | - 7822ee8fe3ec 266 | - ce85771a714c 267 | - db75092f0530 268 | - 5f76965e10cf 269 | - 73fb772e50fb 270 | - def21f50dd3c 271 | - 9c91c546e095 272 | - c535634d7dcd 273 | - 4feda0596965 274 | - 808652a666c6 275 | - 5f40907ec171 276 | - 1f96b9668bdf 277 | - 3665c86afaf5 278 | - 2fc653ca75c7 279 | - a3e59c2ce3f6 280 | -------------------------------------------------------------------------------- /input/folds/stratify_fold_1.yaml: -------------------------------------------------------------------------------- 1 | train_series_ids: 2 | - bfa54bd26187 3 | - d515236bdeec 4 | - 04f547b8017d 5 | - 51c49c540b4e 6 | - 8898e6db816d 7 | - 18b61dd5aae8 8 | - 1319a1935f48 9 | - 1955d568d987 10 | - e34b496b84ce 11 | - 0ef7d94fde99 12 | - 280e08693c6d 13 | - 2e9ced2c7976 14 | - 292a75c0b94e 15 | - f8a8da8bdd00 16 | - 3c336d6ba566 17 | - 89bd631d1769 18 | - 3452b878e596 19 | - b4b75225b224 20 | - 9ee455e4770d 21 | - a4e48102f402 22 | - 3aceb17ef7bd 23 | - e2b60820c325 24 | - b1831c4979da 25 | - 2f7504d0f426 26 | - c7b2155a4a47 27 | - 785c9ca4eff7 28 | - 854206f602d0 29 | - 13b4d6a01d27 30 | - 05e1944c3818 31 | - aed3850f65f0 32 | - 0402a003dae9 33 | - d0f613c700f7 34 | - b7fc34995d0f 35 | - 601559e1777d 36 | - 702bb5387b1e 37 | - cca14d1966c1 38 | - a596ad0b82aa 39 | - 8a306e0890c0 40 | - 91cb6c98201f 41 | - 9a340507e36a 42 | - cfeb11428dd7 43 | - 7822ee8fe3ec 44 | - ce85771a714c 45 | - db75092f0530 46 | - 5f76965e10cf 47 | - 73fb772e50fb 48 | - bfe41e96d12f 49 | - def21f50dd3c 50 | - 9c91c546e095 51 | - c535634d7dcd 52 | - 4feda0596965 53 | - 808652a666c6 54 | - 5f40907ec171 55 | - 1f96b9668bdf 56 | - 29c75c018220 57 | - e2a849d283c0 58 | - 3665c86afaf5 59 | - e0d7b0dcf9f3 60 | - 7504165f497d 61 | - b750c8c1556c 62 | - a88088855de5 63 | - 6ca4f4fca6a2 64 | - 40dce6018935 65 | - 8f6f15b9f598 66 | - 0cfc06c129cc 67 | - 29d3469bd15d 68 | - 1e6717d93c1d 69 | - c289c8a823e0 70 | - 51b23d177971 71 | - dff367373725 72 | - a9a2f7fac455 73 | - 67f5fc60e494 74 | - 3318a0e3ed6f 75 | - dacc6d652e35 76 | - f564985ab692 77 | - 55b7f5c99930 78 | - fa149c3c4bde 79 | - 9ddd40f2cb36 80 | - 2fbbee1a38e3 81 | - bf00506437aa 82 | - 12d01911d509 83 | - c38707ef76df 84 | - 33ceeba8918a 85 | - bb5612895813 86 | - 4ac356361be9 87 | - e0686434d029 88 | - ccdee561ee5d 89 | - 72d2234e84e4 90 | - b737f8c78ec5 91 | - b7188813d58a 92 | - ebd76e93ec7d 93 | - a2b0a64ec9cf 94 | - 60e51cad2ffb 95 | - d3dddd3c0e00 96 | - bccf2f2819f8 97 | - ce9164297046 98 | - 1b92be89db4c 99 | - c7d693f24684 100 | - d5e47b94477e 101 | - 405df1b41f9f 102 | - 939932f1822d 103 | - 0a96f4993bd7 104 | - c8053490cec2 105 | - e1f5abb82285 106 | - 8b8b9e29171c 107 | - eec197a4bdca 108 | - 10469f6765bf 109 | - 3664fe9233f9 110 | - 1716cd4163b2 111 | - e4500e7e19e1 112 | - e30cb792a2bc 113 | - 60d31b0bec3b 114 | - d150801f3145 115 | - 8877a6586606 116 | - 844f54dcab89 117 | - 0f572d690310 118 | - 9277be28a1cf 119 | - 27f09a6a858f 120 | - e1f2a4f991cb 121 | - 655f19eabf1e 122 | - 7fd4284b7ee8 123 | - 612aa8ba44e2 124 | - 0cd1e3d0ed95 125 | - aa81faa78747 126 | - 25e2b3dd9c3b 127 | - 5aad18e7ce64 128 | - 0d0ad1e77851 129 | - 5c088d7e916c 130 | - d043c0ca71cd 131 | - ba8083a2c3b8 132 | - e6ddbaaf0639 133 | - 137b99e936ab 134 | - d8de352c2657 135 | - 653622ac8363 136 | - fbf33b1a2c10 137 | - fb223ed2278c 138 | - dc80ca623d71 139 | - 5e816f11f5c3 140 | - 5f94bb3e1bed 141 | - 483d6545417f 142 | - 3a9a9dc2cbd9 143 | - ad425f3ee76d 144 | - a81f4472c637 145 | - 3df0da2e5966 146 | - a261bc4b7470 147 | - c908a0ad3e31 148 | - 416354edd92a 149 | - 78569a801a38 150 | - 062cae666e2a 151 | - b84960841a75 152 | - d25e479ecbb7 153 | - 10f8bc1f7b07 154 | - 154fe824ed87 155 | - 99237ce045e4 156 | - 6a4cd123bd69 157 | - 1762ab70ec76 158 | - 559ffb7c166a 159 | - 599ca4ed791b 160 | - 971207c6a525 161 | - f0482490923c 162 | - 3be1545083b7 163 | - a167532acca2 164 | - 2654a87be968 165 | - f88e18cb4100 166 | - 9fbdeffbe2ba 167 | - 16fe2798ed0f 168 | - 4ab54be1a403 169 | - c6788e579967 170 | - 137771d19ca2 171 | - 91127c2b0e60 172 | - d5be621fd9aa 173 | - efbfc4526d58 174 | - 2cd2340ca14d 175 | - ee4e0e3afd3d 176 | - 5c55a5e717d6 177 | - 449766346eb1 178 | - f2c2436cf7b7 179 | - af91d9a50547 180 | - f56824b503a0 181 | - 31011ade7c0a 182 | - 6bf95a3cf91c 183 | - 188d4b7cd28b 184 | - e586cbfa7762 185 | - 4a31811f3558 186 | - 927dd0c35dfd 187 | - c5365a55ebb7 188 | - 08db4255286f 189 | - 9b9cd7b7af8c 190 | - a681f9b04b21 191 | - 361366da569e 192 | - 0ec9fc461819 193 | - 44d8c02b369e 194 | - 83fa182bec3a 195 | - 804594bb1f06 196 | - 8becc76ea607 197 | - 519ae2d858b0 198 | - ea0770830757 199 | - e8d0a37c3eba 200 | - 2b8d87addea9 201 | - f7eb179216c2 202 | - 8e32047cbc1f 203 | - c75b4b207bea 204 | - 51fdcc8d9fe7 205 | - 1c7c0bad1263 206 | - 9aed9ee12ae2 207 | - df33ae359fb5 208 | - 87a6cbb7c4ed 209 | - bdfce9ce62b9 210 | - 18a0ca03431d 211 | - 703b5efa9bc1 212 | - 0ce74d6d2106 213 | - 8fb18e36697d 214 | - 148471991ffb 215 | - e69aff66e0cb 216 | - c3072a759efb 217 | - 2fc653ca75c7 218 | - a3e59c2ce3f6 219 | - 0f9e60a8e56d 220 | - 390b487231ce 221 | - e11b9d69f856 222 | - 89c7daa72eee 223 | valid_series_ids: 224 | - 72bbd1ac3edf 225 | - d93b0c7de16b 226 | - 76237b9406d5 227 | - 2b0a1fa8eba8 228 | - 1087d7b0ff2e 229 | - 8b159a98f485 230 | - 8a22387617c3 231 | - 77ca4db83644 232 | - 7476c0bd18d2 233 | - 1d4569cbac0f 234 | - a9e5f5314bcb 235 | - f981a0805fd0 236 | - de6fedfb6139 237 | - 7df249527c63 238 | - d2d6b9af0553 239 | - 694faf956ebf 240 | - b364205aba43 241 | - fcca183903b7 242 | - 3d53bfea61d6 243 | - 55a47ff9dc8a 244 | - 35826366dfc7 245 | - ca732a3c37f7 246 | - fe90110788d2 247 | - d2fef7e4defd 248 | - 72ba4a8afff4 249 | - 4743bdde25df 250 | - ece2561f07e9 251 | - 90eac42a9ec9 252 | - c107b5789660 253 | - 0dee4fda51c3 254 | - 5acc9d63b5fd 255 | - 99b829cbad2d 256 | - db5e0ee1c0ab 257 | - 4b45c36f8f5a 258 | - e867b5133665 259 | - f6d2cc003183 260 | - 752900afe3a6 261 | - eef041dd50aa 262 | - 349c5562ee2c 263 | - 5ffd5e1e81ac 264 | - dfc3ccebfdc9 265 | - d9e887091a5c 266 | - 062dbd4c95e6 267 | - 3be2f86c3e45 268 | - c68260cc9e8f 269 | - ebb6fae8ed43 270 | - cf13ed7e457a 271 | - 03d92c9f6f8a 272 | - 6d6b9d22d48a 273 | - ca730dbf521d 274 | - 038441c925bb 275 | - 44a41bba1ee7 276 | - 207eded97727 277 | - 6ee4ade1f2bd 278 | - c5d08fc3e040 279 | - c7b1283bb7eb 280 | -------------------------------------------------------------------------------- /input/folds/stratify_fold_2.yaml: -------------------------------------------------------------------------------- 1 | train_series_ids: 2 | - bfa54bd26187 3 | - d515236bdeec 4 | - 04f547b8017d 5 | - 51c49c540b4e 6 | - 8898e6db816d 7 | - 18b61dd5aae8 8 | - 1319a1935f48 9 | - 1955d568d987 10 | - e34b496b84ce 11 | - 0ef7d94fde99 12 | - 280e08693c6d 13 | - 2e9ced2c7976 14 | - 292a75c0b94e 15 | - f8a8da8bdd00 16 | - 3c336d6ba566 17 | - 89bd631d1769 18 | - 3452b878e596 19 | - b4b75225b224 20 | - 9ee455e4770d 21 | - a4e48102f402 22 | - 3aceb17ef7bd 23 | - e2b60820c325 24 | - b1831c4979da 25 | - 2f7504d0f426 26 | - c7b2155a4a47 27 | - 785c9ca4eff7 28 | - 854206f602d0 29 | - 13b4d6a01d27 30 | - 72bbd1ac3edf 31 | - 05e1944c3818 32 | - aed3850f65f0 33 | - 0402a003dae9 34 | - d0f613c700f7 35 | - b7fc34995d0f 36 | - d93b0c7de16b 37 | - 601559e1777d 38 | - 702bb5387b1e 39 | - cca14d1966c1 40 | - a596ad0b82aa 41 | - 8a306e0890c0 42 | - 91cb6c98201f 43 | - 9a340507e36a 44 | - cfeb11428dd7 45 | - 7822ee8fe3ec 46 | - 76237b9406d5 47 | - ce85771a714c 48 | - db75092f0530 49 | - 2b0a1fa8eba8 50 | - 5f76965e10cf 51 | - 1087d7b0ff2e 52 | - 8b159a98f485 53 | - 73fb772e50fb 54 | - 8a22387617c3 55 | - def21f50dd3c 56 | - 9c91c546e095 57 | - c535634d7dcd 58 | - 77ca4db83644 59 | - 7476c0bd18d2 60 | - 1d4569cbac0f 61 | - 4feda0596965 62 | - 808652a666c6 63 | - a9e5f5314bcb 64 | - 5f40907ec171 65 | - 1f96b9668bdf 66 | - f981a0805fd0 67 | - de6fedfb6139 68 | - 7df249527c63 69 | - d2d6b9af0553 70 | - 694faf956ebf 71 | - b364205aba43 72 | - fcca183903b7 73 | - 3d53bfea61d6 74 | - 55a47ff9dc8a 75 | - 35826366dfc7 76 | - ca732a3c37f7 77 | - fe90110788d2 78 | - d2fef7e4defd 79 | - 72ba4a8afff4 80 | - 4743bdde25df 81 | - ece2561f07e9 82 | - 90eac42a9ec9 83 | - c107b5789660 84 | - 0dee4fda51c3 85 | - 5acc9d63b5fd 86 | - 99b829cbad2d 87 | - db5e0ee1c0ab 88 | - 4b45c36f8f5a 89 | - e867b5133665 90 | - f6d2cc003183 91 | - 752900afe3a6 92 | - eef041dd50aa 93 | - 349c5562ee2c 94 | - 5ffd5e1e81ac 95 | - 3665c86afaf5 96 | - dfc3ccebfdc9 97 | - d9e887091a5c 98 | - 062dbd4c95e6 99 | - 3be2f86c3e45 100 | - c68260cc9e8f 101 | - ebb6fae8ed43 102 | - cf13ed7e457a 103 | - 03d92c9f6f8a 104 | - 6d6b9d22d48a 105 | - ca730dbf521d 106 | - 038441c925bb 107 | - 44a41bba1ee7 108 | - 207eded97727 109 | - 6ee4ade1f2bd 110 | - b737f8c78ec5 111 | - d3dddd3c0e00 112 | - ce9164297046 113 | - 939932f1822d 114 | - 0a96f4993bd7 115 | - e1f5abb82285 116 | - eec197a4bdca 117 | - 3664fe9233f9 118 | - 1716cd4163b2 119 | - e4500e7e19e1 120 | - 8877a6586606 121 | - 844f54dcab89 122 | - 0f572d690310 123 | - 9277be28a1cf 124 | - 27f09a6a858f 125 | - 655f19eabf1e 126 | - 612aa8ba44e2 127 | - aa81faa78747 128 | - 25e2b3dd9c3b 129 | - 5aad18e7ce64 130 | - 0d0ad1e77851 131 | - 5c088d7e916c 132 | - d043c0ca71cd 133 | - ba8083a2c3b8 134 | - e6ddbaaf0639 135 | - d8de352c2657 136 | - 653622ac8363 137 | - fbf33b1a2c10 138 | - fb223ed2278c 139 | - dc80ca623d71 140 | - 5e816f11f5c3 141 | - 483d6545417f 142 | - 3a9a9dc2cbd9 143 | - ad425f3ee76d 144 | - a81f4472c637 145 | - 3df0da2e5966 146 | - a261bc4b7470 147 | - c908a0ad3e31 148 | - 416354edd92a 149 | - 78569a801a38 150 | - 062cae666e2a 151 | - b84960841a75 152 | - d25e479ecbb7 153 | - 10f8bc1f7b07 154 | - 154fe824ed87 155 | - 99237ce045e4 156 | - 6a4cd123bd69 157 | - 1762ab70ec76 158 | - 559ffb7c166a 159 | - 599ca4ed791b 160 | - 971207c6a525 161 | - f0482490923c 162 | - 3be1545083b7 163 | - a167532acca2 164 | - 2654a87be968 165 | - f88e18cb4100 166 | - 9fbdeffbe2ba 167 | - 16fe2798ed0f 168 | - 4ab54be1a403 169 | - c6788e579967 170 | - 137771d19ca2 171 | - 91127c2b0e60 172 | - d5be621fd9aa 173 | - efbfc4526d58 174 | - 2cd2340ca14d 175 | - ee4e0e3afd3d 176 | - 5c55a5e717d6 177 | - 449766346eb1 178 | - f2c2436cf7b7 179 | - af91d9a50547 180 | - f56824b503a0 181 | - 31011ade7c0a 182 | - 6bf95a3cf91c 183 | - 188d4b7cd28b 184 | - e586cbfa7762 185 | - 4a31811f3558 186 | - 927dd0c35dfd 187 | - c5365a55ebb7 188 | - 08db4255286f 189 | - 9b9cd7b7af8c 190 | - a681f9b04b21 191 | - 361366da569e 192 | - 0ec9fc461819 193 | - 44d8c02b369e 194 | - 83fa182bec3a 195 | - 804594bb1f06 196 | - 8becc76ea607 197 | - 519ae2d858b0 198 | - ea0770830757 199 | - e8d0a37c3eba 200 | - 2b8d87addea9 201 | - f7eb179216c2 202 | - 8e32047cbc1f 203 | - c75b4b207bea 204 | - 51fdcc8d9fe7 205 | - 1c7c0bad1263 206 | - 9aed9ee12ae2 207 | - df33ae359fb5 208 | - 87a6cbb7c4ed 209 | - bdfce9ce62b9 210 | - 18a0ca03431d 211 | - 703b5efa9bc1 212 | - 0ce74d6d2106 213 | - 8fb18e36697d 214 | - 148471991ffb 215 | - e69aff66e0cb 216 | - c3072a759efb 217 | - 2fc653ca75c7 218 | - a3e59c2ce3f6 219 | - c5d08fc3e040 220 | - c7b1283bb7eb 221 | - 390b487231ce 222 | - e11b9d69f856 223 | - 89c7daa72eee 224 | valid_series_ids: 225 | - bfe41e96d12f 226 | - 29c75c018220 227 | - e2a849d283c0 228 | - e0d7b0dcf9f3 229 | - 7504165f497d 230 | - b750c8c1556c 231 | - a88088855de5 232 | - 6ca4f4fca6a2 233 | - 40dce6018935 234 | - 8f6f15b9f598 235 | - 0cfc06c129cc 236 | - 29d3469bd15d 237 | - 1e6717d93c1d 238 | - c289c8a823e0 239 | - 51b23d177971 240 | - dff367373725 241 | - a9a2f7fac455 242 | - 67f5fc60e494 243 | - 3318a0e3ed6f 244 | - dacc6d652e35 245 | - f564985ab692 246 | - 55b7f5c99930 247 | - fa149c3c4bde 248 | - 9ddd40f2cb36 249 | - 2fbbee1a38e3 250 | - bf00506437aa 251 | - 12d01911d509 252 | - c38707ef76df 253 | - 33ceeba8918a 254 | - bb5612895813 255 | - 4ac356361be9 256 | - e0686434d029 257 | - ccdee561ee5d 258 | - 72d2234e84e4 259 | - b7188813d58a 260 | - ebd76e93ec7d 261 | - a2b0a64ec9cf 262 | - 60e51cad2ffb 263 | - bccf2f2819f8 264 | - 1b92be89db4c 265 | - c7d693f24684 266 | - d5e47b94477e 267 | - 405df1b41f9f 268 | - c8053490cec2 269 | - 8b8b9e29171c 270 | - 10469f6765bf 271 | - e30cb792a2bc 272 | - 60d31b0bec3b 273 | - d150801f3145 274 | - e1f2a4f991cb 275 | - 7fd4284b7ee8 276 | - 0cd1e3d0ed95 277 | - 137b99e936ab 278 | - 5f94bb3e1bed 279 | - 0f9e60a8e56d 280 | -------------------------------------------------------------------------------- /input/folds/stratify_fold_3.yaml: -------------------------------------------------------------------------------- 1 | train_series_ids: 2 | - bfa54bd26187 3 | - d515236bdeec 4 | - 04f547b8017d 5 | - 51c49c540b4e 6 | - 8898e6db816d 7 | - 18b61dd5aae8 8 | - 1319a1935f48 9 | - 1955d568d987 10 | - e34b496b84ce 11 | - 0ef7d94fde99 12 | - 280e08693c6d 13 | - 2e9ced2c7976 14 | - 292a75c0b94e 15 | - f8a8da8bdd00 16 | - 3c336d6ba566 17 | - 89bd631d1769 18 | - 3452b878e596 19 | - b4b75225b224 20 | - 9ee455e4770d 21 | - a4e48102f402 22 | - 3aceb17ef7bd 23 | - e2b60820c325 24 | - b1831c4979da 25 | - 2f7504d0f426 26 | - c7b2155a4a47 27 | - 785c9ca4eff7 28 | - 854206f602d0 29 | - 13b4d6a01d27 30 | - 72bbd1ac3edf 31 | - 05e1944c3818 32 | - aed3850f65f0 33 | - 0402a003dae9 34 | - d0f613c700f7 35 | - b7fc34995d0f 36 | - d93b0c7de16b 37 | - 601559e1777d 38 | - 702bb5387b1e 39 | - cca14d1966c1 40 | - a596ad0b82aa 41 | - 8a306e0890c0 42 | - 91cb6c98201f 43 | - 9a340507e36a 44 | - cfeb11428dd7 45 | - 7822ee8fe3ec 46 | - 76237b9406d5 47 | - ce85771a714c 48 | - db75092f0530 49 | - 2b0a1fa8eba8 50 | - 5f76965e10cf 51 | - 1087d7b0ff2e 52 | - 8b159a98f485 53 | - 73fb772e50fb 54 | - bfe41e96d12f 55 | - 8a22387617c3 56 | - def21f50dd3c 57 | - 9c91c546e095 58 | - c535634d7dcd 59 | - 77ca4db83644 60 | - 7476c0bd18d2 61 | - 1d4569cbac0f 62 | - 4feda0596965 63 | - 808652a666c6 64 | - a9e5f5314bcb 65 | - 5f40907ec171 66 | - 1f96b9668bdf 67 | - f981a0805fd0 68 | - de6fedfb6139 69 | - 7df249527c63 70 | - d2d6b9af0553 71 | - 694faf956ebf 72 | - b364205aba43 73 | - fcca183903b7 74 | - 3d53bfea61d6 75 | - 55a47ff9dc8a 76 | - 35826366dfc7 77 | - ca732a3c37f7 78 | - fe90110788d2 79 | - d2fef7e4defd 80 | - 72ba4a8afff4 81 | - 4743bdde25df 82 | - ece2561f07e9 83 | - 90eac42a9ec9 84 | - c107b5789660 85 | - 0dee4fda51c3 86 | - 5acc9d63b5fd 87 | - 99b829cbad2d 88 | - db5e0ee1c0ab 89 | - 4b45c36f8f5a 90 | - e867b5133665 91 | - f6d2cc003183 92 | - 752900afe3a6 93 | - eef041dd50aa 94 | - 349c5562ee2c 95 | - 5ffd5e1e81ac 96 | - 29c75c018220 97 | - e2a849d283c0 98 | - 3665c86afaf5 99 | - dfc3ccebfdc9 100 | - d9e887091a5c 101 | - 062dbd4c95e6 102 | - e0d7b0dcf9f3 103 | - 3be2f86c3e45 104 | - c68260cc9e8f 105 | - ebb6fae8ed43 106 | - cf13ed7e457a 107 | - 03d92c9f6f8a 108 | - 6d6b9d22d48a 109 | - ca730dbf521d 110 | - 7504165f497d 111 | - b750c8c1556c 112 | - 038441c925bb 113 | - 44a41bba1ee7 114 | - a88088855de5 115 | - 6ca4f4fca6a2 116 | - 40dce6018935 117 | - 8f6f15b9f598 118 | - 0cfc06c129cc 119 | - 29d3469bd15d 120 | - 1e6717d93c1d 121 | - c289c8a823e0 122 | - 51b23d177971 123 | - dff367373725 124 | - a9a2f7fac455 125 | - 67f5fc60e494 126 | - 3318a0e3ed6f 127 | - dacc6d652e35 128 | - 207eded97727 129 | - f564985ab692 130 | - 55b7f5c99930 131 | - fa149c3c4bde 132 | - 9ddd40f2cb36 133 | - 2fbbee1a38e3 134 | - bf00506437aa 135 | - 12d01911d509 136 | - c38707ef76df 137 | - 33ceeba8918a 138 | - bb5612895813 139 | - 4ac356361be9 140 | - e0686434d029 141 | - 6ee4ade1f2bd 142 | - ccdee561ee5d 143 | - 72d2234e84e4 144 | - b7188813d58a 145 | - ebd76e93ec7d 146 | - a2b0a64ec9cf 147 | - 60e51cad2ffb 148 | - bccf2f2819f8 149 | - 1b92be89db4c 150 | - c7d693f24684 151 | - d5e47b94477e 152 | - 405df1b41f9f 153 | - c8053490cec2 154 | - 8b8b9e29171c 155 | - 10469f6765bf 156 | - e30cb792a2bc 157 | - 60d31b0bec3b 158 | - d150801f3145 159 | - e1f2a4f991cb 160 | - 7fd4284b7ee8 161 | - 0cd1e3d0ed95 162 | - 137b99e936ab 163 | - 5f94bb3e1bed 164 | - 416354edd92a 165 | - 062cae666e2a 166 | - 6a4cd123bd69 167 | - 559ffb7c166a 168 | - 971207c6a525 169 | - 3be1545083b7 170 | - f88e18cb4100 171 | - 137771d19ca2 172 | - d5be621fd9aa 173 | - efbfc4526d58 174 | - 2cd2340ca14d 175 | - ee4e0e3afd3d 176 | - 5c55a5e717d6 177 | - 449766346eb1 178 | - f2c2436cf7b7 179 | - af91d9a50547 180 | - f56824b503a0 181 | - 31011ade7c0a 182 | - 6bf95a3cf91c 183 | - 188d4b7cd28b 184 | - e586cbfa7762 185 | - 4a31811f3558 186 | - 927dd0c35dfd 187 | - c5365a55ebb7 188 | - 08db4255286f 189 | - 9b9cd7b7af8c 190 | - a681f9b04b21 191 | - 361366da569e 192 | - 0ec9fc461819 193 | - 44d8c02b369e 194 | - 83fa182bec3a 195 | - 804594bb1f06 196 | - 8becc76ea607 197 | - 519ae2d858b0 198 | - ea0770830757 199 | - e8d0a37c3eba 200 | - 2b8d87addea9 201 | - f7eb179216c2 202 | - 8e32047cbc1f 203 | - c75b4b207bea 204 | - 51fdcc8d9fe7 205 | - 1c7c0bad1263 206 | - 9aed9ee12ae2 207 | - df33ae359fb5 208 | - 87a6cbb7c4ed 209 | - bdfce9ce62b9 210 | - 18a0ca03431d 211 | - 703b5efa9bc1 212 | - 0ce74d6d2106 213 | - 8fb18e36697d 214 | - 148471991ffb 215 | - e69aff66e0cb 216 | - c3072a759efb 217 | - 2fc653ca75c7 218 | - a3e59c2ce3f6 219 | - c5d08fc3e040 220 | - c7b1283bb7eb 221 | - 0f9e60a8e56d 222 | - e11b9d69f856 223 | - 89c7daa72eee 224 | valid_series_ids: 225 | - b737f8c78ec5 226 | - d3dddd3c0e00 227 | - ce9164297046 228 | - 939932f1822d 229 | - 0a96f4993bd7 230 | - e1f5abb82285 231 | - eec197a4bdca 232 | - 3664fe9233f9 233 | - 1716cd4163b2 234 | - e4500e7e19e1 235 | - 8877a6586606 236 | - 844f54dcab89 237 | - 0f572d690310 238 | - 9277be28a1cf 239 | - 27f09a6a858f 240 | - 655f19eabf1e 241 | - 612aa8ba44e2 242 | - aa81faa78747 243 | - 25e2b3dd9c3b 244 | - 5aad18e7ce64 245 | - 0d0ad1e77851 246 | - 5c088d7e916c 247 | - d043c0ca71cd 248 | - ba8083a2c3b8 249 | - e6ddbaaf0639 250 | - d8de352c2657 251 | - 653622ac8363 252 | - fbf33b1a2c10 253 | - fb223ed2278c 254 | - dc80ca623d71 255 | - 5e816f11f5c3 256 | - 483d6545417f 257 | - 3a9a9dc2cbd9 258 | - ad425f3ee76d 259 | - a81f4472c637 260 | - 3df0da2e5966 261 | - a261bc4b7470 262 | - c908a0ad3e31 263 | - 78569a801a38 264 | - b84960841a75 265 | - d25e479ecbb7 266 | - 10f8bc1f7b07 267 | - 154fe824ed87 268 | - 99237ce045e4 269 | - 1762ab70ec76 270 | - 599ca4ed791b 271 | - f0482490923c 272 | - a167532acca2 273 | - 2654a87be968 274 | - 9fbdeffbe2ba 275 | - 16fe2798ed0f 276 | - 4ab54be1a403 277 | - c6788e579967 278 | - 91127c2b0e60 279 | - 390b487231ce 280 | -------------------------------------------------------------------------------- /input/folds/stratify_fold_4.yaml: -------------------------------------------------------------------------------- 1 | train_series_ids: 2 | - bfa54bd26187 3 | - d515236bdeec 4 | - 04f547b8017d 5 | - 51c49c540b4e 6 | - 8898e6db816d 7 | - 18b61dd5aae8 8 | - 1319a1935f48 9 | - 1955d568d987 10 | - e34b496b84ce 11 | - 0ef7d94fde99 12 | - 280e08693c6d 13 | - 2e9ced2c7976 14 | - 292a75c0b94e 15 | - f8a8da8bdd00 16 | - 3c336d6ba566 17 | - 89bd631d1769 18 | - 3452b878e596 19 | - b4b75225b224 20 | - 9ee455e4770d 21 | - a4e48102f402 22 | - 3aceb17ef7bd 23 | - e2b60820c325 24 | - b1831c4979da 25 | - 2f7504d0f426 26 | - c7b2155a4a47 27 | - 785c9ca4eff7 28 | - 854206f602d0 29 | - 13b4d6a01d27 30 | - 72bbd1ac3edf 31 | - 05e1944c3818 32 | - aed3850f65f0 33 | - 0402a003dae9 34 | - d0f613c700f7 35 | - b7fc34995d0f 36 | - d93b0c7de16b 37 | - 601559e1777d 38 | - 702bb5387b1e 39 | - cca14d1966c1 40 | - a596ad0b82aa 41 | - 8a306e0890c0 42 | - 91cb6c98201f 43 | - 9a340507e36a 44 | - cfeb11428dd7 45 | - 7822ee8fe3ec 46 | - 76237b9406d5 47 | - ce85771a714c 48 | - db75092f0530 49 | - 2b0a1fa8eba8 50 | - 5f76965e10cf 51 | - 1087d7b0ff2e 52 | - 8b159a98f485 53 | - 73fb772e50fb 54 | - bfe41e96d12f 55 | - 8a22387617c3 56 | - def21f50dd3c 57 | - 9c91c546e095 58 | - c535634d7dcd 59 | - 77ca4db83644 60 | - 7476c0bd18d2 61 | - 1d4569cbac0f 62 | - 4feda0596965 63 | - 808652a666c6 64 | - a9e5f5314bcb 65 | - 5f40907ec171 66 | - 1f96b9668bdf 67 | - f981a0805fd0 68 | - de6fedfb6139 69 | - 7df249527c63 70 | - d2d6b9af0553 71 | - 694faf956ebf 72 | - b364205aba43 73 | - fcca183903b7 74 | - 3d53bfea61d6 75 | - 55a47ff9dc8a 76 | - 35826366dfc7 77 | - ca732a3c37f7 78 | - fe90110788d2 79 | - d2fef7e4defd 80 | - 72ba4a8afff4 81 | - 4743bdde25df 82 | - ece2561f07e9 83 | - 90eac42a9ec9 84 | - c107b5789660 85 | - 0dee4fda51c3 86 | - 5acc9d63b5fd 87 | - 99b829cbad2d 88 | - db5e0ee1c0ab 89 | - 4b45c36f8f5a 90 | - e867b5133665 91 | - f6d2cc003183 92 | - 752900afe3a6 93 | - eef041dd50aa 94 | - 349c5562ee2c 95 | - 5ffd5e1e81ac 96 | - 29c75c018220 97 | - e2a849d283c0 98 | - 3665c86afaf5 99 | - dfc3ccebfdc9 100 | - d9e887091a5c 101 | - 062dbd4c95e6 102 | - e0d7b0dcf9f3 103 | - 3be2f86c3e45 104 | - c68260cc9e8f 105 | - ebb6fae8ed43 106 | - cf13ed7e457a 107 | - 03d92c9f6f8a 108 | - 6d6b9d22d48a 109 | - ca730dbf521d 110 | - 7504165f497d 111 | - b750c8c1556c 112 | - 038441c925bb 113 | - 44a41bba1ee7 114 | - a88088855de5 115 | - 6ca4f4fca6a2 116 | - 40dce6018935 117 | - 8f6f15b9f598 118 | - 0cfc06c129cc 119 | - 29d3469bd15d 120 | - 1e6717d93c1d 121 | - c289c8a823e0 122 | - 51b23d177971 123 | - dff367373725 124 | - a9a2f7fac455 125 | - 67f5fc60e494 126 | - 3318a0e3ed6f 127 | - dacc6d652e35 128 | - 207eded97727 129 | - f564985ab692 130 | - 55b7f5c99930 131 | - fa149c3c4bde 132 | - 9ddd40f2cb36 133 | - 2fbbee1a38e3 134 | - bf00506437aa 135 | - 12d01911d509 136 | - c38707ef76df 137 | - 33ceeba8918a 138 | - bb5612895813 139 | - 4ac356361be9 140 | - e0686434d029 141 | - 6ee4ade1f2bd 142 | - ccdee561ee5d 143 | - 72d2234e84e4 144 | - b737f8c78ec5 145 | - b7188813d58a 146 | - ebd76e93ec7d 147 | - a2b0a64ec9cf 148 | - 60e51cad2ffb 149 | - d3dddd3c0e00 150 | - bccf2f2819f8 151 | - ce9164297046 152 | - 1b92be89db4c 153 | - c7d693f24684 154 | - d5e47b94477e 155 | - 405df1b41f9f 156 | - 939932f1822d 157 | - 0a96f4993bd7 158 | - c8053490cec2 159 | - e1f5abb82285 160 | - 8b8b9e29171c 161 | - eec197a4bdca 162 | - 10469f6765bf 163 | - 3664fe9233f9 164 | - 1716cd4163b2 165 | - e4500e7e19e1 166 | - e30cb792a2bc 167 | - 60d31b0bec3b 168 | - d150801f3145 169 | - 8877a6586606 170 | - 844f54dcab89 171 | - 0f572d690310 172 | - 9277be28a1cf 173 | - 27f09a6a858f 174 | - e1f2a4f991cb 175 | - 655f19eabf1e 176 | - 7fd4284b7ee8 177 | - 612aa8ba44e2 178 | - 0cd1e3d0ed95 179 | - aa81faa78747 180 | - 25e2b3dd9c3b 181 | - 5aad18e7ce64 182 | - 0d0ad1e77851 183 | - 5c088d7e916c 184 | - d043c0ca71cd 185 | - ba8083a2c3b8 186 | - e6ddbaaf0639 187 | - 137b99e936ab 188 | - d8de352c2657 189 | - 653622ac8363 190 | - fbf33b1a2c10 191 | - fb223ed2278c 192 | - dc80ca623d71 193 | - 5e816f11f5c3 194 | - 5f94bb3e1bed 195 | - 483d6545417f 196 | - 3a9a9dc2cbd9 197 | - ad425f3ee76d 198 | - a81f4472c637 199 | - 3df0da2e5966 200 | - a261bc4b7470 201 | - c908a0ad3e31 202 | - 78569a801a38 203 | - b84960841a75 204 | - d25e479ecbb7 205 | - 10f8bc1f7b07 206 | - 154fe824ed87 207 | - 99237ce045e4 208 | - 1762ab70ec76 209 | - 599ca4ed791b 210 | - f0482490923c 211 | - a167532acca2 212 | - 2654a87be968 213 | - 9fbdeffbe2ba 214 | - 16fe2798ed0f 215 | - 4ab54be1a403 216 | - c6788e579967 217 | - 91127c2b0e60 218 | - 2fc653ca75c7 219 | - a3e59c2ce3f6 220 | - c5d08fc3e040 221 | - c7b1283bb7eb 222 | - 0f9e60a8e56d 223 | - 390b487231ce 224 | valid_series_ids: 225 | - 416354edd92a 226 | - 062cae666e2a 227 | - 6a4cd123bd69 228 | - 559ffb7c166a 229 | - 971207c6a525 230 | - 3be1545083b7 231 | - f88e18cb4100 232 | - 137771d19ca2 233 | - d5be621fd9aa 234 | - efbfc4526d58 235 | - 2cd2340ca14d 236 | - ee4e0e3afd3d 237 | - 5c55a5e717d6 238 | - 449766346eb1 239 | - f2c2436cf7b7 240 | - af91d9a50547 241 | - f56824b503a0 242 | - 31011ade7c0a 243 | - 6bf95a3cf91c 244 | - 188d4b7cd28b 245 | - e586cbfa7762 246 | - 4a31811f3558 247 | - 927dd0c35dfd 248 | - c5365a55ebb7 249 | - 08db4255286f 250 | - 9b9cd7b7af8c 251 | - a681f9b04b21 252 | - 361366da569e 253 | - 0ec9fc461819 254 | - 44d8c02b369e 255 | - 83fa182bec3a 256 | - 804594bb1f06 257 | - 8becc76ea607 258 | - 519ae2d858b0 259 | - ea0770830757 260 | - e8d0a37c3eba 261 | - 2b8d87addea9 262 | - f7eb179216c2 263 | - 8e32047cbc1f 264 | - c75b4b207bea 265 | - 51fdcc8d9fe7 266 | - 1c7c0bad1263 267 | - 9aed9ee12ae2 268 | - df33ae359fb5 269 | - 87a6cbb7c4ed 270 | - bdfce9ce62b9 271 | - 18a0ca03431d 272 | - 703b5efa9bc1 273 | - 0ce74d6d2106 274 | - 8fb18e36697d 275 | - 148471991ffb 276 | - e69aff66e0cb 277 | - c3072a759efb 278 | - e11b9d69f856 279 | - 89c7daa72eee 280 | -------------------------------------------------------------------------------- /kami/run/conf/split/stratify_fold_0.yaml: -------------------------------------------------------------------------------- 1 | train_series_ids: 2 | - 72bbd1ac3edf 3 | - d93b0c7de16b 4 | - 76237b9406d5 5 | - 2b0a1fa8eba8 6 | - 1087d7b0ff2e 7 | - 8b159a98f485 8 | - bfe41e96d12f 9 | - 8a22387617c3 10 | - 77ca4db83644 11 | - 7476c0bd18d2 12 | - 1d4569cbac0f 13 | - a9e5f5314bcb 14 | - f981a0805fd0 15 | - de6fedfb6139 16 | - 7df249527c63 17 | - d2d6b9af0553 18 | - 694faf956ebf 19 | - b364205aba43 20 | - fcca183903b7 21 | - 3d53bfea61d6 22 | - 55a47ff9dc8a 23 | - 35826366dfc7 24 | - ca732a3c37f7 25 | - fe90110788d2 26 | - d2fef7e4defd 27 | - 72ba4a8afff4 28 | - 4743bdde25df 29 | - ece2561f07e9 30 | - 90eac42a9ec9 31 | - c107b5789660 32 | - 0dee4fda51c3 33 | - 5acc9d63b5fd 34 | - 99b829cbad2d 35 | - db5e0ee1c0ab 36 | - 4b45c36f8f5a 37 | - e867b5133665 38 | - f6d2cc003183 39 | - 752900afe3a6 40 | - eef041dd50aa 41 | - 349c5562ee2c 42 | - 5ffd5e1e81ac 43 | - 29c75c018220 44 | - e2a849d283c0 45 | - dfc3ccebfdc9 46 | - d9e887091a5c 47 | - 062dbd4c95e6 48 | - e0d7b0dcf9f3 49 | - 3be2f86c3e45 50 | - c68260cc9e8f 51 | - ebb6fae8ed43 52 | - cf13ed7e457a 53 | - 03d92c9f6f8a 54 | - 6d6b9d22d48a 55 | - ca730dbf521d 56 | - 7504165f497d 57 | - b750c8c1556c 58 | - 038441c925bb 59 | - 44a41bba1ee7 60 | - a88088855de5 61 | - 6ca4f4fca6a2 62 | - 40dce6018935 63 | - 8f6f15b9f598 64 | - 0cfc06c129cc 65 | - 29d3469bd15d 66 | - 1e6717d93c1d 67 | - c289c8a823e0 68 | - 51b23d177971 69 | - dff367373725 70 | - a9a2f7fac455 71 | - 67f5fc60e494 72 | - 3318a0e3ed6f 73 | - dacc6d652e35 74 | - 207eded97727 75 | - f564985ab692 76 | - 55b7f5c99930 77 | - fa149c3c4bde 78 | - 9ddd40f2cb36 79 | - 2fbbee1a38e3 80 | - bf00506437aa 81 | - 12d01911d509 82 | - c38707ef76df 83 | - 33ceeba8918a 84 | - bb5612895813 85 | - 4ac356361be9 86 | - e0686434d029 87 | - 6ee4ade1f2bd 88 | - ccdee561ee5d 89 | - 72d2234e84e4 90 | - b737f8c78ec5 91 | - b7188813d58a 92 | - ebd76e93ec7d 93 | - a2b0a64ec9cf 94 | - 60e51cad2ffb 95 | - d3dddd3c0e00 96 | - bccf2f2819f8 97 | - ce9164297046 98 | - 1b92be89db4c 99 | - c7d693f24684 100 | - d5e47b94477e 101 | - 405df1b41f9f 102 | - 939932f1822d 103 | - 0a96f4993bd7 104 | - c8053490cec2 105 | - e1f5abb82285 106 | - 8b8b9e29171c 107 | - eec197a4bdca 108 | - 10469f6765bf 109 | - 3664fe9233f9 110 | - 1716cd4163b2 111 | - e4500e7e19e1 112 | - e30cb792a2bc 113 | - 60d31b0bec3b 114 | - d150801f3145 115 | - 8877a6586606 116 | - 844f54dcab89 117 | - 0f572d690310 118 | - 9277be28a1cf 119 | - 27f09a6a858f 120 | - e1f2a4f991cb 121 | - 655f19eabf1e 122 | - 7fd4284b7ee8 123 | - 612aa8ba44e2 124 | - 0cd1e3d0ed95 125 | - aa81faa78747 126 | - 25e2b3dd9c3b 127 | - 5aad18e7ce64 128 | - 0d0ad1e77851 129 | - 5c088d7e916c 130 | - d043c0ca71cd 131 | - ba8083a2c3b8 132 | - e6ddbaaf0639 133 | - 137b99e936ab 134 | - d8de352c2657 135 | - 653622ac8363 136 | - fbf33b1a2c10 137 | - fb223ed2278c 138 | - dc80ca623d71 139 | - 5e816f11f5c3 140 | - 5f94bb3e1bed 141 | - 483d6545417f 142 | - 3a9a9dc2cbd9 143 | - ad425f3ee76d 144 | - a81f4472c637 145 | - 3df0da2e5966 146 | - a261bc4b7470 147 | - c908a0ad3e31 148 | - 416354edd92a 149 | - 78569a801a38 150 | - 062cae666e2a 151 | - b84960841a75 152 | - d25e479ecbb7 153 | - 10f8bc1f7b07 154 | - 154fe824ed87 155 | - 99237ce045e4 156 | - 6a4cd123bd69 157 | - 1762ab70ec76 158 | - 559ffb7c166a 159 | - 599ca4ed791b 160 | - 971207c6a525 161 | - f0482490923c 162 | - 3be1545083b7 163 | - a167532acca2 164 | - 2654a87be968 165 | - f88e18cb4100 166 | - 9fbdeffbe2ba 167 | - 16fe2798ed0f 168 | - 4ab54be1a403 169 | - c6788e579967 170 | - 137771d19ca2 171 | - 91127c2b0e60 172 | - d5be621fd9aa 173 | - efbfc4526d58 174 | - 2cd2340ca14d 175 | - ee4e0e3afd3d 176 | - 5c55a5e717d6 177 | - 449766346eb1 178 | - f2c2436cf7b7 179 | - af91d9a50547 180 | - f56824b503a0 181 | - 31011ade7c0a 182 | - 6bf95a3cf91c 183 | - 188d4b7cd28b 184 | - e586cbfa7762 185 | - 4a31811f3558 186 | - 927dd0c35dfd 187 | - c5365a55ebb7 188 | - 08db4255286f 189 | - 9b9cd7b7af8c 190 | - a681f9b04b21 191 | - 361366da569e 192 | - 0ec9fc461819 193 | - 44d8c02b369e 194 | - 83fa182bec3a 195 | - 804594bb1f06 196 | - 8becc76ea607 197 | - 519ae2d858b0 198 | - ea0770830757 199 | - e8d0a37c3eba 200 | - 2b8d87addea9 201 | - f7eb179216c2 202 | - 8e32047cbc1f 203 | - c75b4b207bea 204 | - 51fdcc8d9fe7 205 | - 1c7c0bad1263 206 | - 9aed9ee12ae2 207 | - df33ae359fb5 208 | - 87a6cbb7c4ed 209 | - bdfce9ce62b9 210 | - 18a0ca03431d 211 | - 703b5efa9bc1 212 | - 0ce74d6d2106 213 | - 8fb18e36697d 214 | - 148471991ffb 215 | - e69aff66e0cb 216 | - c3072a759efb 217 | - c5d08fc3e040 218 | - c7b1283bb7eb 219 | - 0f9e60a8e56d 220 | - 390b487231ce 221 | - e11b9d69f856 222 | - 89c7daa72eee 223 | valid_series_ids: 224 | - bfa54bd26187 225 | - d515236bdeec 226 | - 04f547b8017d 227 | - 51c49c540b4e 228 | - 8898e6db816d 229 | - 18b61dd5aae8 230 | - 1319a1935f48 231 | - 1955d568d987 232 | - e34b496b84ce 233 | - 0ef7d94fde99 234 | - 280e08693c6d 235 | - 2e9ced2c7976 236 | - 292a75c0b94e 237 | - f8a8da8bdd00 238 | - 3c336d6ba566 239 | - 89bd631d1769 240 | - 3452b878e596 241 | - b4b75225b224 242 | - 9ee455e4770d 243 | - a4e48102f402 244 | - 3aceb17ef7bd 245 | - e2b60820c325 246 | - b1831c4979da 247 | - 2f7504d0f426 248 | - c7b2155a4a47 249 | - 785c9ca4eff7 250 | - 854206f602d0 251 | - 13b4d6a01d27 252 | - 05e1944c3818 253 | - aed3850f65f0 254 | - 0402a003dae9 255 | - d0f613c700f7 256 | - b7fc34995d0f 257 | - 601559e1777d 258 | - 702bb5387b1e 259 | - cca14d1966c1 260 | - a596ad0b82aa 261 | - 8a306e0890c0 262 | - 91cb6c98201f 263 | - 9a340507e36a 264 | - cfeb11428dd7 265 | - 7822ee8fe3ec 266 | - ce85771a714c 267 | - db75092f0530 268 | - 5f76965e10cf 269 | - 73fb772e50fb 270 | - def21f50dd3c 271 | - 9c91c546e095 272 | - c535634d7dcd 273 | - 4feda0596965 274 | - 808652a666c6 275 | - 5f40907ec171 276 | - 1f96b9668bdf 277 | - 3665c86afaf5 278 | - 2fc653ca75c7 279 | - a3e59c2ce3f6 280 | -------------------------------------------------------------------------------- /kami/run/conf/split/stratify_fold_1.yaml: -------------------------------------------------------------------------------- 1 | train_series_ids: 2 | - bfa54bd26187 3 | - d515236bdeec 4 | - 04f547b8017d 5 | - 51c49c540b4e 6 | - 8898e6db816d 7 | - 18b61dd5aae8 8 | - 1319a1935f48 9 | - 1955d568d987 10 | - e34b496b84ce 11 | - 0ef7d94fde99 12 | - 280e08693c6d 13 | - 2e9ced2c7976 14 | - 292a75c0b94e 15 | - f8a8da8bdd00 16 | - 3c336d6ba566 17 | - 89bd631d1769 18 | - 3452b878e596 19 | - b4b75225b224 20 | - 9ee455e4770d 21 | - a4e48102f402 22 | - 3aceb17ef7bd 23 | - e2b60820c325 24 | - b1831c4979da 25 | - 2f7504d0f426 26 | - c7b2155a4a47 27 | - 785c9ca4eff7 28 | - 854206f602d0 29 | - 13b4d6a01d27 30 | - 05e1944c3818 31 | - aed3850f65f0 32 | - 0402a003dae9 33 | - d0f613c700f7 34 | - b7fc34995d0f 35 | - 601559e1777d 36 | - 702bb5387b1e 37 | - cca14d1966c1 38 | - a596ad0b82aa 39 | - 8a306e0890c0 40 | - 91cb6c98201f 41 | - 9a340507e36a 42 | - cfeb11428dd7 43 | - 7822ee8fe3ec 44 | - ce85771a714c 45 | - db75092f0530 46 | - 5f76965e10cf 47 | - 73fb772e50fb 48 | - bfe41e96d12f 49 | - def21f50dd3c 50 | - 9c91c546e095 51 | - c535634d7dcd 52 | - 4feda0596965 53 | - 808652a666c6 54 | - 5f40907ec171 55 | - 1f96b9668bdf 56 | - 29c75c018220 57 | - e2a849d283c0 58 | - 3665c86afaf5 59 | - e0d7b0dcf9f3 60 | - 7504165f497d 61 | - b750c8c1556c 62 | - a88088855de5 63 | - 6ca4f4fca6a2 64 | - 40dce6018935 65 | - 8f6f15b9f598 66 | - 0cfc06c129cc 67 | - 29d3469bd15d 68 | - 1e6717d93c1d 69 | - c289c8a823e0 70 | - 51b23d177971 71 | - dff367373725 72 | - a9a2f7fac455 73 | - 67f5fc60e494 74 | - 3318a0e3ed6f 75 | - dacc6d652e35 76 | - f564985ab692 77 | - 55b7f5c99930 78 | - fa149c3c4bde 79 | - 9ddd40f2cb36 80 | - 2fbbee1a38e3 81 | - bf00506437aa 82 | - 12d01911d509 83 | - c38707ef76df 84 | - 33ceeba8918a 85 | - bb5612895813 86 | - 4ac356361be9 87 | - e0686434d029 88 | - ccdee561ee5d 89 | - 72d2234e84e4 90 | - b737f8c78ec5 91 | - b7188813d58a 92 | - ebd76e93ec7d 93 | - a2b0a64ec9cf 94 | - 60e51cad2ffb 95 | - d3dddd3c0e00 96 | - bccf2f2819f8 97 | - ce9164297046 98 | - 1b92be89db4c 99 | - c7d693f24684 100 | - d5e47b94477e 101 | - 405df1b41f9f 102 | - 939932f1822d 103 | - 0a96f4993bd7 104 | - c8053490cec2 105 | - e1f5abb82285 106 | - 8b8b9e29171c 107 | - eec197a4bdca 108 | - 10469f6765bf 109 | - 3664fe9233f9 110 | - 1716cd4163b2 111 | - e4500e7e19e1 112 | - e30cb792a2bc 113 | - 60d31b0bec3b 114 | - d150801f3145 115 | - 8877a6586606 116 | - 844f54dcab89 117 | - 0f572d690310 118 | - 9277be28a1cf 119 | - 27f09a6a858f 120 | - e1f2a4f991cb 121 | - 655f19eabf1e 122 | - 7fd4284b7ee8 123 | - 612aa8ba44e2 124 | - 0cd1e3d0ed95 125 | - aa81faa78747 126 | - 25e2b3dd9c3b 127 | - 5aad18e7ce64 128 | - 0d0ad1e77851 129 | - 5c088d7e916c 130 | - d043c0ca71cd 131 | - ba8083a2c3b8 132 | - e6ddbaaf0639 133 | - 137b99e936ab 134 | - d8de352c2657 135 | - 653622ac8363 136 | - fbf33b1a2c10 137 | - fb223ed2278c 138 | - dc80ca623d71 139 | - 5e816f11f5c3 140 | - 5f94bb3e1bed 141 | - 483d6545417f 142 | - 3a9a9dc2cbd9 143 | - ad425f3ee76d 144 | - a81f4472c637 145 | - 3df0da2e5966 146 | - a261bc4b7470 147 | - c908a0ad3e31 148 | - 416354edd92a 149 | - 78569a801a38 150 | - 062cae666e2a 151 | - b84960841a75 152 | - d25e479ecbb7 153 | - 10f8bc1f7b07 154 | - 154fe824ed87 155 | - 99237ce045e4 156 | - 6a4cd123bd69 157 | - 1762ab70ec76 158 | - 559ffb7c166a 159 | - 599ca4ed791b 160 | - 971207c6a525 161 | - f0482490923c 162 | - 3be1545083b7 163 | - a167532acca2 164 | - 2654a87be968 165 | - f88e18cb4100 166 | - 9fbdeffbe2ba 167 | - 16fe2798ed0f 168 | - 4ab54be1a403 169 | - c6788e579967 170 | - 137771d19ca2 171 | - 91127c2b0e60 172 | - d5be621fd9aa 173 | - efbfc4526d58 174 | - 2cd2340ca14d 175 | - ee4e0e3afd3d 176 | - 5c55a5e717d6 177 | - 449766346eb1 178 | - f2c2436cf7b7 179 | - af91d9a50547 180 | - f56824b503a0 181 | - 31011ade7c0a 182 | - 6bf95a3cf91c 183 | - 188d4b7cd28b 184 | - e586cbfa7762 185 | - 4a31811f3558 186 | - 927dd0c35dfd 187 | - c5365a55ebb7 188 | - 08db4255286f 189 | - 9b9cd7b7af8c 190 | - a681f9b04b21 191 | - 361366da569e 192 | - 0ec9fc461819 193 | - 44d8c02b369e 194 | - 83fa182bec3a 195 | - 804594bb1f06 196 | - 8becc76ea607 197 | - 519ae2d858b0 198 | - ea0770830757 199 | - e8d0a37c3eba 200 | - 2b8d87addea9 201 | - f7eb179216c2 202 | - 8e32047cbc1f 203 | - c75b4b207bea 204 | - 51fdcc8d9fe7 205 | - 1c7c0bad1263 206 | - 9aed9ee12ae2 207 | - df33ae359fb5 208 | - 87a6cbb7c4ed 209 | - bdfce9ce62b9 210 | - 18a0ca03431d 211 | - 703b5efa9bc1 212 | - 0ce74d6d2106 213 | - 8fb18e36697d 214 | - 148471991ffb 215 | - e69aff66e0cb 216 | - c3072a759efb 217 | - 2fc653ca75c7 218 | - a3e59c2ce3f6 219 | - 0f9e60a8e56d 220 | - 390b487231ce 221 | - e11b9d69f856 222 | - 89c7daa72eee 223 | valid_series_ids: 224 | - 72bbd1ac3edf 225 | - d93b0c7de16b 226 | - 76237b9406d5 227 | - 2b0a1fa8eba8 228 | - 1087d7b0ff2e 229 | - 8b159a98f485 230 | - 8a22387617c3 231 | - 77ca4db83644 232 | - 7476c0bd18d2 233 | - 1d4569cbac0f 234 | - a9e5f5314bcb 235 | - f981a0805fd0 236 | - de6fedfb6139 237 | - 7df249527c63 238 | - d2d6b9af0553 239 | - 694faf956ebf 240 | - b364205aba43 241 | - fcca183903b7 242 | - 3d53bfea61d6 243 | - 55a47ff9dc8a 244 | - 35826366dfc7 245 | - ca732a3c37f7 246 | - fe90110788d2 247 | - d2fef7e4defd 248 | - 72ba4a8afff4 249 | - 4743bdde25df 250 | - ece2561f07e9 251 | - 90eac42a9ec9 252 | - c107b5789660 253 | - 0dee4fda51c3 254 | - 5acc9d63b5fd 255 | - 99b829cbad2d 256 | - db5e0ee1c0ab 257 | - 4b45c36f8f5a 258 | - e867b5133665 259 | - f6d2cc003183 260 | - 752900afe3a6 261 | - eef041dd50aa 262 | - 349c5562ee2c 263 | - 5ffd5e1e81ac 264 | - dfc3ccebfdc9 265 | - d9e887091a5c 266 | - 062dbd4c95e6 267 | - 3be2f86c3e45 268 | - c68260cc9e8f 269 | - ebb6fae8ed43 270 | - cf13ed7e457a 271 | - 03d92c9f6f8a 272 | - 6d6b9d22d48a 273 | - ca730dbf521d 274 | - 038441c925bb 275 | - 44a41bba1ee7 276 | - 207eded97727 277 | - 6ee4ade1f2bd 278 | - c5d08fc3e040 279 | - c7b1283bb7eb 280 | --------------------------------------------------------------------------------