├── LICENSE
├── README.md
├── dlt2.xls
├── dlt4all.py
├── dlt4all_test.py
├── hyper_parameters.py
├── poems
    ├── model.py
    ├── poems.py
    └── resnet.py
├── ssq.py
├── ssq.xls
├── ssq4all.py
├── ssq4all_test.py
├── ssq4all_test_v4.py
├── ssq4all_v4.py
├── ssq_data.py
└── ssq_test.py


/LICENSE:
--------------------------------------------------------------------------------
 1 | MIT License
 2 | 
 3 | Copyright (c) 2018 Yanchao Liang
 4 | 
 5 | Permission is hereby granted, free of charge, to any person obtaining a copy
 6 | of this software and associated documentation files (the "Software"), to deal
 7 | in the Software without restriction, including without limitation the rights
 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 9 | copies of the Software, and to permit persons to whom the Software is
10 | furnished to do so, subject to the following conditions:
11 | 
12 | The above copyright notice and this permission notice shall be included in all
13 | copies or substantial portions of the Software.
14 | 
15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 | SOFTWARE.
22 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # 利用神经网络和LSTM预测双色球(How To predict the China's Union Lotto with Neural Network and LSTM)
 2 | ## 2020/12/4更新
 3 | 因为问的人比较多了，所以提供了适配'tensorflow 2.X'版本的[分支](https://github.com/Liang-yc/ssq/tree/tf2).
 4 | 
 5 | ## 2019/6/18更新
 6 | 利用最近几次的双色球结果和奖池情况做预测，训练和测试见`ssq4all_v4.py`和`ssq4all_test_v4.py`。
 7 | ## 2018/9/16更新
 8 | 添加了大乐透的训练测试文件。
 9 | ## 2018/5/30更新
10 | 根据反馈，修复了一些错误，删除了两个文件。另外，有问题请直接开issue。
11 | ## 2018/3/29更新
12 | 尝试用CNN和LSTM做预测。CNN用于提取特征，采用的是resnet。目前最好的结果是五等奖（中4个红号）。
13 | ## 背景(Background)
14 | ------
15 | 这个项目是通过结合神经网络和Long Short-Term Memory(LSTM)完成双色球预测。关于双色球的介绍，在此处不在赘述（参见[网站](https://baike.baidu.com/item/中国福利彩票双色球/8676030?fr=aladdin&fromid=75279&fromtitle=%E5%8F%8C%E8%89%B2%E7%90%83)）。该项目以真实的双色球开奖结果作为输入（7个数值，其中前6个表示红号，最后一个表示蓝号），输出预测结果（输出仍为7个数值，其中前6个表示红号，最后一个表示蓝号）。目前该项目还处于开发中。<br><br>
16 | 其中，项目核心代码主要借鉴了这个[网站](https://github.com/jinfagang/tensorflow_poems) 。
17 | 截至目前，该项目充分训练后的预测结果中，正确预测的红号数目为0-2个（1个的居多，但鉴于这概率本身就偏高，因此不能算是取得很好的结果），正确预测的蓝号数目为0-1个（0个的居多），考虑到双色球的中奖条件（至少3红号正确或者至少蓝号正确），**不建议各位直接根据预测结果去买**。<br>
18 | 预测结果不好有以下几个原因：
19 | 1. 双色球本身是一个随机事件，而我却希望从中找到规律；
20 | 2. 数据集样本数量不够多且无法有效扩大数据集；
21 | 3. 可能我的模型还不够优秀。
22 | 
23 | 如果您有什么好的建议或者有什么问题，请直接发我邮箱（yc_liang@qq.com）或者在issue留言也可以；<br>
24 | 如果您觉得这项目似乎还有点意思，记得STAR；<br>
25 | 当然，如果您真的中奖了，要求不高, STAR。<br>
26 | 
27 | ## 项目依赖(Requirement)
28 | -----
29 | 
30 | 1.Tensorflow;<br>
31 | 2.pyexcel_xls;<br>
32 | 3.CUDA.(optional)<br>
33 | 该项目的代码可能还需要一些python包但我没列出来，请自行pip一下。<br><br>
34 | 
35 | ## 项目文件(File contents)
36 | -----
37 | 
38 | 1.` poems` : ` poems`文件夹包含3个模型文件，其中`model.py`是本项目所使用的模型文件;其中`resnet.py`定义的是resnet模型;<br>
39 | 2.`ssq.py `: 该文件用于训练双色球模型;<br>
40 | 3.`ssq.xls` : 该文件储存了历次双色球的开奖和中奖数据，需要利用宏进行数据更新;<br>
41 | 4.`ssq_data.py ` : 读取`ssq.xls`文件中的数据并变成所需要的形式.<br>
42 | 5.`ssq_test.py`: 以最近一次的双色球开奖结果作为输入，输出模型产生的预测结果;<br>
43 | 6.`ssq4all.py `: 该文件用于训练双色球模型（和`ssq.py `功能一样，但用的模型不一样）;<br>
44 | 7.`ssq4all_test.py`: 以最近一次的双色球开奖结果作为输入，输出模型产生的预测结果;<br>
45 | 6.`dlt.xls` : 该文件储存了历次双色球的开奖和中奖数据，需要利用宏进行数据更新;<br>
46 | 7.`dlt4all.py `: 该文件用于训练大乐透模型;<br>
47 | 8.`dlt4all_test.py`: 以最近一次的大乐透开奖结果作为输入，输出模型产生的预测结果;<br>
48 | 
49 | ## 训练及测试(How To Use)
50 | -----
51 | 设置好参数，运行ssq.py（建议在控制台运行，这样训练到一半可以直接通过按CTRL+C取消训练并保存结果；这样下次训练时可以从上一次的checkpoint继续训练）。训练后，项目目录下会生成一个model文件夹用于保存模型。然后运行ssq_test.py进行预测就可以了。<br>
52 | 该项目在W10, tensorflow 1.4/1.5正常运行。有无GPU均可运行。原则上讲，这项目的模型并不大，且数据量也不大，所以可以直接将所有训练数据作为一个batch进行训练，后续如果需要加深模型，可能要降低batch size。
53 | 


--------------------------------------------------------------------------------
/dlt2.xls:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Liang-yc/ssq/4c69745d1fb731b4a39cc571af58a31368439789/dlt2.xls


--------------------------------------------------------------------------------
/dlt4all.py:
--------------------------------------------------------------------------------
  1 | # -*- coding: utf-8 -*-
  2 | # file: main.py
  3 | # author: JinTian
  4 | # time: 11/03/2017 9:53 AM
  5 | # Copyright 2017 JinTian. All Rights Reserved.
  6 | #
  7 | # Licensed under the Apache License, Version 2.0 (the "License");
  8 | # you may not use this file except in compliance with the License.
  9 | # You may obtain a copy of the License at
 10 | #
 11 | #     http://www.apache.org/licenses/LICENSE-2.0
 12 | #
 13 | # Unless required by applicable law or agreed to in writing, software
 14 | # distributed under the License is distributed on an "AS IS" BASIS,
 15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 16 | # See the License for the specific language governing permissions and
 17 | # limitations under the License.
 18 | # ------------------------------------------------------------------------
 19 | import os
 20 | import numpy as np
 21 | import tensorflow as tf
 22 | from poems.model import rnn_model
 23 | from poems.resnet import *
 24 | from poems.poems import process_poems, generate_batch
 25 | from ssq_data import *
 26 | # for Windows10：OSError: raw write() returned invalid length 96 (should have been between 0 and 48)
 27 | import win_unicode_console
 28 | win_unicode_console.enable()
 29 | tf.app.flags.DEFINE_integer('batch_size', 2214, 'batch size.')
 30 | tf.app.flags.DEFINE_float('learning_rate', 0.0001, 'learning rate.')
 31 | tf.app.flags.DEFINE_string('model_dir', os.path.abspath('./dlt_model'), 'model save path.')
 32 | tf.app.flags.DEFINE_string('file_path', os.path.abspath('./data/poems.txt'), 'file name of poems.')
 33 | tf.app.flags.DEFINE_string('model_prefix', 'poems', 'model save prefix.')
 34 | tf.app.flags.DEFINE_integer('epochs', 500000, 'train how many epochs.')
 35 | 
 36 | FLAGS = tf.app.flags.FLAGS
 37 | 
 38 | 
 39 | def run_training():
 40 |     if not os.path.exists(FLAGS.model_dir):
 41 |         os.makedirs(FLAGS.model_dir)
 42 |     #
 43 |     # poems_vector, word_to_int, vocabularies = process_poems(FLAGS.file_path)
 44 |     # batches_inputs, batches_outputs = generate_batch(FLAGS.batch_size, poems_vector, word_to_int)
 45 |     # ssqdata=get_exl_data(random_order=False,use_resnet=True)
 46 |     # # print(ssqdata[len(ssqdata)-1])
 47 |     # batches_inputs=ssqdata[0:(len(ssqdata)-1)]
 48 |     # ssqdata=get_exl_data(random_order=False,use_resnet=False)
 49 |     ssqdata=get_dlt_data(random_order=False,use_resnet=True)
 50 |     # print(ssqdata[len(ssqdata)-1])
 51 |     batches_inputs=ssqdata[0:(len(ssqdata)-1)]
 52 |     ssqdata=get_dlt_data(random_order=False,use_resnet=False)
 53 |     batches_outputs = ssqdata[1:(len(ssqdata))]
 54 |     FLAGS.batch_size=len(batches_inputs)
 55 |     # print(np.shape(batches_outputs))
 56 |     # data=batches_outputs[1:7]
 57 |     # print(len(data))
 58 |     del ssqdata
 59 |     input_data = tf.placeholder(tf.float32, [FLAGS.batch_size, 1,7,1])
 60 |     logits = inference(input_data, 1, reuse=False,output_num=128)
 61 | 
 62 |     # print(tf.shape(input_data))
 63 |     output_targets = tf.placeholder(tf.int32, [FLAGS.batch_size, None])
 64 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=35+12,output_num=7,
 65 |                            rnn_size=128, num_layers=7, batch_size=FLAGS.batch_size, learning_rate=FLAGS.learning_rate)
 66 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=output_targets, vocab_size=len(
 67 |     #     vocabularies), rnn_size=128, num_layers=2, batch_size=64, learning_rate=FLAGS.learning_rate)
 68 | 
 69 |     saver = tf.train.Saver(tf.global_variables())
 70 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
 71 |     with tf.Session() as sess:
 72 |         # sess = tf_debug.LocalCLIDebugWrapperSession(sess=sess)
 73 |         # sess.add_tensor_filter("has_inf_or_nan", tf_debug.has_inf_or_nan)
 74 |         sess.run(init_op)
 75 | 
 76 |         start_epoch = 0
 77 |         # saver.restore(sess, "D:/tensorflow_poems-master/model4all/poems-207261")
 78 |         checkpoint = tf.train.latest_checkpoint(FLAGS.model_dir)
 79 |         if checkpoint:
 80 |             saver.restore(sess, checkpoint)
 81 |             print("## restore from the checkpoint {0}".format(checkpoint))
 82 |             start_epoch += int(checkpoint.split('-')[-1])
 83 |         print('## start training...')
 84 |         try:
 85 |             for epoch in range(start_epoch, FLAGS.epochs):
 86 |                 n = 0
 87 |                 # n_chunk = len(poems_vector) // FLAGS.batch_size
 88 |                 # n_chunk = len(batches_inputs) // FLAGS.batch_size
 89 |                 n_chunk=math.ceil(len(batches_inputs) / FLAGS.batch_size)
 90 |                 for batch in range(n_chunk):
 91 |                     left=(batch+1)*FLAGS.batch_size-len(batches_inputs)
 92 |                     if left<0:
 93 |                         inputdata=batches_inputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 94 |                         outputdata=batches_outputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 95 |                     else:
 96 |                         # temp=batches_inputs[batch*FLAGS.batch_size:len(batches_inputs) ]
 97 |                         # temp.extend(batches_inputs[0:left])
 98 |                         inputdata=batches_inputs[len(batches_inputs)-FLAGS.batch_size:len(batches_inputs)]
 99 |                         # temp=batches_outputs[batch*FLAGS.batch_size:len(batches_inputs) ]
100 |                         # temp.extend(batches_outputs[0:left])
101 |                         outputdata=batches_outputs[len(batches_outputs)-FLAGS.batch_size:len(batches_outputs)]
102 |                     # print(len(inputdata))
103 |                     loss, _, _ = sess.run([
104 |                         end_points['total_loss'],
105 |                         end_points['last_state'],
106 |                         end_points['train_op']
107 |                     ], feed_dict={input_data: inputdata, output_targets: outputdata})
108 |                     # ], feed_dict={input_data: batches_inputs, output_targets: batches_outputs})
109 |                     n += 1
110 |                 if epoch % 1000 == 0:
111 |                     print('Epoch: %d, batch: %d, training loss: %.6f' % (epoch, batch, loss))
112 |                 if epoch % 50000 == 0:
113 |                     saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
114 |         except KeyboardInterrupt:
115 |             print('## Interrupt manually, try saving checkpoint for now...')
116 |         finally:
117 |             saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
118 |             print('## Last epoch were saved, next time will start from epoch {}.'.format(epoch))
119 | 
120 | 
121 | def main(_):
122 |     run_training()
123 | 
124 | 
125 | if __name__ == '__main__':
126 |     tf.app.run()
127 | 


--------------------------------------------------------------------------------
/dlt4all_test.py:
--------------------------------------------------------------------------------
  1 | # -*- coding: utf-8 -*-
  2 | # file: main.py
  3 | # author: JinTian
  4 | # time: 11/03/2017 9:53 AM
  5 | # Copyright 2017 JinTian. All Rights Reserved.
  6 | #
  7 | # Licensed under the Apache License, Version 2.0 (the "License");
  8 | # you may not use this file except in compliance with the License.
  9 | # You may obtain a copy of the License at
 10 | #
 11 | #     http://www.apache.org/licenses/LICENSE-2.0
 12 | #
 13 | # Unless required by applicable law or agreed to in writing, software
 14 | # distributed under the License is distributed on an "AS IS" BASIS,
 15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 16 | # See the License for the specific language governing permissions and
 17 | # limitations under the License.
 18 | # ------------------------------------------------------------------------
 19 | import tensorflow as tf
 20 | from poems.model import rnn_model
 21 | from poems.resnet import *
 22 | from poems.poems import process_poems
 23 | import numpy as np
 24 | from ssq_data import *
 25 | start_token = 'B'
 26 | end_token = 'E'
 27 | model_dir = './model4all/'
 28 | corpus_file = './data/poems.txt'
 29 | 
 30 | 
 31 | def to_word(predict, vocabs):
 32 |     t = np.cumsum(predict)
 33 |     s = np.sum(predict)
 34 |     sample = int(np.searchsorted(t, np.random.rand(1) * s))
 35 |     if sample > len(vocabs):
 36 |         sample = len(vocabs) - 1
 37 |     return vocabs[sample]
 38 | 
 39 | 
 40 | def gen_poem():
 41 |     batch_size = 1
 42 |     print('## loading model from %s' % model_dir)
 43 |     input_data = tf.placeholder(tf.float32, [1, 1,7,1])
 44 |     logits = inference(input_data, 1, reuse=False,output_num=128)
 45 | 
 46 |     # print(tf.shape(input_data))
 47 |     output_targets = tf.placeholder(tf.int32, [1, None])
 48 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=35+12,
 49 |                            output_num=7,
 50 |                            rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.01)
 51 | 
 52 |     # input_data = tf.placeholder(tf.int32, [batch_size, None])
 53 |     #
 54 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=None, vocab_size=33,
 55 |     #                        rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.01)
 56 | 
 57 |     saver = tf.train.Saver(tf.global_variables())
 58 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
 59 |     with tf.Session() as sess:
 60 |         sess.run(init_op)
 61 | 
 62 |         checkpoint = tf.train.latest_checkpoint('./dlt_model/')
 63 |         saver.restore(sess, checkpoint)
 64 |         # saver.restore(sess, "E:/workplace/tensorflow_poems-master/model4red/poems-150000")
 65 |         ssqdata = get_dlt_data(random_order=False,use_resnet=True)
 66 |         # x = np.array([list(map(word_int_map.get, start_token))])
 67 |         x=[ssqdata[len(ssqdata)-1]]
 68 |         print("input: %s"%(x+np.asarray([[[[1],[1],[1],[1],[1],[-34],[-34]]]])))
 69 |         [predict, last_state] = sess.run([end_points['prediction'], end_points['last_state']],
 70 |                                          feed_dict={input_data: x})
 71 |         poem_=np.argmax(np.array(predict),axis=1)
 72 |         sorted_result = np.argsort(np.array(predict), axis=1)
 73 |         results=poem_+np.asarray([1,1,1,1,1,-34,-34])
 74 |         # print(sorted_result)
 75 |         print("output: %s"%results)
 76 |         return poem_
 77 | 
 78 | def gen_blue():
 79 |     batch_size = 1
 80 |     print('## loading model from %s' % model_dir)
 81 |     input_data = tf.placeholder(tf.float32, [1, 1,7,1])
 82 |     logits = inference(input_data, 10, reuse=False,output_num=128)
 83 | 
 84 |     # print(tf.shape(input_data))
 85 |     output_targets = tf.placeholder(tf.int32, [1, None])
 86 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=33,
 87 |                            output_num=1,
 88 |                            rnn_size=128, num_layers=3, batch_size=1, learning_rate=0.01)
 89 | 
 90 |     # input_data = tf.placeholder(tf.int32, [batch_size, None])
 91 |     #
 92 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=None, vocab_size=33,
 93 |     #                        rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.01)
 94 | 
 95 |     saver = tf.train.Saver(tf.global_variables())
 96 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
 97 |     with tf.Session() as sess:
 98 |         sess.run(init_op)
 99 | 
100 |         checkpoint = tf.train.latest_checkpoint('./model4all/')
101 |         saver.restore(sess, checkpoint)
102 |         # saver.restore(sess, "E:/workplace/tensorflow_poems-master/model4blue/poems-84201")
103 |         ssqdata = get_exl_data(random_order=False,use_resnet=True)
104 |         # x = np.array([list(map(word_int_map.get, start_token))])
105 |         x=[ssqdata[len(ssqdata)-1]]
106 |         print("input: %s"%(x))
107 |         [predict, last_state] = sess.run([end_points['prediction'], end_points['last_state']],
108 |                                          feed_dict={input_data: x})
109 |         poem_=np.argmax(np.array(predict),axis=1)
110 |         sorted_result=np.argsort(np.array(predict),axis=1)
111 |         results=poem_+np.ones(1)
112 |         print(results)
113 |         print(sorted_result)
114 |         return poem_
115 | 
116 | 
117 | if __name__ == '__main__':
118 |     # begin_char = input('## please input the first character:')
119 |     # poem=gen_blue()#13
120 |     poem = gen_poem()
121 |     # pretty_print_poem(poem_=poem)


--------------------------------------------------------------------------------
/hyper_parameters.py:
--------------------------------------------------------------------------------
 1 | # Coder: Wenxin Xu
 2 | # Github: https://github.com/wenxinxu/resnet_in_tensorflow
 3 | # ==============================================================================
 4 | import tensorflow as tf
 5 | 
 6 | FLAGS = tf.app.flags.FLAGS
 7 | 
 8 | ## The following flags are related to save paths, tensorboard outputs and screen outputs
 9 | 
10 | tf.app.flags.DEFINE_string('version', 'test_110', '''A version number defining the directory to save
11 | logs and checkpoints''')
12 | tf.app.flags.DEFINE_integer('report_freq', 391, '''Steps takes to output errors on the screen
13 | and write summaries''')
14 | tf.app.flags.DEFINE_float('train_ema_decay', 0.95, '''The decay factor of the train error's
15 | moving average shown on tensorboard''')
16 | 
17 | 
18 | ## The following flags define hyper-parameters regards training
19 | 
20 | tf.app.flags.DEFINE_integer('train_steps', 80000, '''Total steps that you want to train''')
21 | tf.app.flags.DEFINE_boolean('is_full_validation', False, '''Validation w/ full validation set or
22 | a random batch''')
23 | tf.app.flags.DEFINE_integer('train_batch_size', 128, '''Train batch size''')
24 | tf.app.flags.DEFINE_integer('validation_batch_size', 250, '''Validation batch size, better to be
25 | a divisor of 10000 for this task''')
26 | tf.app.flags.DEFINE_integer('test_batch_size', 125, '''Test batch size''')
27 | 
28 | tf.app.flags.DEFINE_float('init_lr', 0.1, '''Initial learning rate''')
29 | tf.app.flags.DEFINE_float('lr_decay_factor', 0.1, '''How much to decay the learning rate each
30 | time''')
31 | tf.app.flags.DEFINE_integer('decay_step0', 40000, '''At which step to decay the learning rate''')
32 | tf.app.flags.DEFINE_integer('decay_step1', 60000, '''At which step to decay the learning rate''')
33 | 
34 | 
35 | ## The following flags define hyper-parameters modifying the training network
36 | 
37 | tf.app.flags.DEFINE_integer('num_residual_blocks', 5, '''How many residual blocks do you want''')
38 | tf.app.flags.DEFINE_float('weight_decay', 0.0002, '''scale for l2 regularization''')
39 | 
40 | 
41 | ## The following flags are related to data-augmentation
42 | 
43 | tf.app.flags.DEFINE_integer('padding_size', 2, '''In data augmentation, layers of zero padding on
44 | each side of the image''')
45 | 
46 | 
47 | ## If you want to load a checkpoint and continue training
48 | 
49 | tf.app.flags.DEFINE_string('ckpt_path', 'cache/logs_repeat20/model.ckpt-100000', '''Checkpoint
50 | directory to restore''')
51 | tf.app.flags.DEFINE_boolean('is_use_ckpt', False, '''Whether to load a checkpoint and continue
52 | training''')
53 | 
54 | tf.app.flags.DEFINE_string('test_ckpt_path', 'model_110.ckpt-79999', '''Checkpoint
55 | directory to restore''')
56 | 
57 | 
58 | train_dir = 'logs_' + FLAGS.version + '/'
59 | 


--------------------------------------------------------------------------------
/poems/model.py:
--------------------------------------------------------------------------------
  1 | # -*- coding: utf-8 -*-
  2 | # file: model.py
  3 | # author: JinTian
  4 | # time: 07/03/2017 3:07 PM
  5 | # Copyright 2017 JinTian. All Rights Reserved.
  6 | #
  7 | # Licensed under the Apache License, Version 2.0 (the "License");
  8 | # you may not use this file except in compliance with the License.
  9 | # You may obtain a copy of the License at
 10 | #
 11 | #     http://www.apache.org/licenses/LICENSE-2.0
 12 | #
 13 | # Unless required by applicable law or agreed to in writing, software
 14 | # distributed under the License is distributed on an "AS IS" BASIS,
 15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 16 | # See the License for the specific language governing permissions and
 17 | # limitations under the License.
 18 | # ------------------------------------------------------------------------
 19 | import tensorflow as tf
 20 | import numpy as np
 21 | 
 22 | 
 23 | def rnn_model(model, input_data, output_data, vocab_size, rnn_size=128, num_layers=2, batch_size=64,
 24 |               input_num=128,output_num=1,learning_rate=0.01,use_cnn=True):
 25 |     """
 26 |     construct rnn seq2seq model.
 27 |     :param model: model class
 28 |     :param input_data: input data placeholder
 29 |     :param output_data: output data placeholder
 30 |     :param vocab_size:
 31 |     :param rnn_size:
 32 |     :param num_layers:
 33 |     :param batch_size:
 34 |     :param learning_rate:
 35 |     :return:
 36 |     """
 37 |     end_points = {}
 38 | 
 39 | 
 40 |     if model == 'rnn':
 41 |         cell_fun = tf.contrib.rnn.BasicRNNCell
 42 |         cell = cell_fun(rnn_size)
 43 |     elif model == 'gru':
 44 |         cell_fun = tf.contrib.rnn.GRUCell
 45 |         cell = cell_fun(rnn_size)
 46 |     elif model == 'lstm':
 47 |         cell_fun = tf.contrib.rnn.BasicLSTMCell
 48 |         cell = cell_fun(rnn_size, state_is_tuple=True)
 49 |         
 50 |     cell = tf.contrib.rnn.MultiRNNCell([cell] * num_layers, state_is_tuple=True)
 51 | 
 52 |     if output_data is not None:
 53 |         initial_state = cell.zero_state(batch_size, tf.float32)
 54 |     else:
 55 |         initial_state = cell.zero_state(1, tf.float32)
 56 |     if use_cnn:
 57 |         with tf.name_scope('fc1'):
 58 |             fc1_weights = tf.Variable(  # fully connected, depth 512.
 59 |                 tf.truncated_normal([input_num, 128],
 60 |                                     # mean=1.0,
 61 |                                     stddev=0.1,
 62 |                                     dtype=tf.float32))
 63 |             fc1_biases = tf.Variable(tf.constant(1.0, shape=[128], dtype=tf.float32))
 64 |             fc1 = tf.nn.relu(tf.matmul(tf.to_float(input_data), fc1_weights) + fc1_biases)
 65 |         with tf.name_scope('fc2'):
 66 |             fc2_weights = tf.Variable(  # fully connected, depth 512.
 67 |                 tf.truncated_normal([128, 256],
 68 |                                     # mean=1.0,
 69 |                                     stddev=0.1,
 70 |                                     dtype=tf.float32))
 71 |             fc2_biases = tf.Variable(tf.constant(1.0, shape=[256], dtype=tf.float32))
 72 |             fc2 = tf.nn.relu(tf.matmul(tf.to_float(fc1), fc2_weights) + fc2_biases)
 73 |         with tf.name_scope('fc3'):
 74 |             fc3_weights = tf.Variable(  # fully connected, depth 512.
 75 |                 tf.truncated_normal([256, 512],
 76 |                                     # mean=1.0,
 77 |                                     stddev=0.1,
 78 |                                     dtype=tf.float32))
 79 |             fc3_biases = tf.Variable(tf.constant(1.0, shape=[512], dtype=tf.float32))
 80 |             fc3 = tf.nn.relu(tf.matmul(tf.to_float(fc2), fc3_weights) + fc3_biases)
 81 |         with tf.name_scope('fc4'):
 82 |             fc4_weights = tf.Variable(  # fully connected, depth 512.
 83 |                 tf.truncated_normal([512, 128*output_num],
 84 |                                     # mean=1.0,
 85 |                                     stddev=0.1,
 86 |                                     dtype=tf.float32))
 87 |             fc4_biases = tf.Variable(tf.constant(1.0, shape=[128*output_num], dtype=tf.float32))
 88 |             fc4 = tf.nn.relu(tf.matmul(fc3, fc4_weights) + fc4_biases)
 89 | 
 90 | 
 91 |         # [batch_size, ?, rnn_size] = [64, ?, 128]
 92 |         embedded=tf.reshape(fc4,[batch_size,-1,128])
 93 |         outputs, last_state = tf.nn.dynamic_rnn(cell, embedded, initial_state=initial_state)
 94 |     else:
 95 |         with tf.device("/cpu:0"):
 96 |             embedding = tf.get_variable('embedding', initializer=tf.random_uniform(
 97 |                 [vocab_size + 1, rnn_size], -1.0, 1.0))
 98 |             inputs = tf.nn.embedding_lookup(embedding, input_data)
 99 |     # embedded = tf.reshape(input_data, [batch_size, -1, 128])
100 |     # outputs, last_state = tf.nn.dynamic_rnn(cell, embedded, initial_state=initial_state)
101 |     # outputs, last_state = tf.nn.dynamic_rnn(cell, inputs, initial_state=initial_state)
102 |     output = tf.reshape(outputs, [-1, rnn_size])
103 | 
104 |     weights = tf.Variable(tf.truncated_normal([rnn_size, vocab_size + 1]))
105 |     bias = tf.Variable(tf.zeros(shape=[vocab_size + 1]))
106 |     logits = tf.nn.bias_add(tf.matmul(output, weights), bias=bias)
107 |     # [?, vocab_size+1]
108 | 
109 |     if output_data is not None:
110 |         # output_data must be one-hot encode
111 |         labels = tf.one_hot(tf.reshape(output_data, [-1]), depth=vocab_size + 1)
112 |         # should be [?, vocab_size+1]
113 |         loss = tf.nn.softmax_cross_entropy_with_logits(labels=labels, logits=logits)
114 | 
115 |         # loss shape should be [?, vocab_size+1]
116 |         total_loss = tf.reduce_mean(loss)
117 |         regularizers = (tf.nn.l2_loss(fc1_weights) + tf.nn.l2_loss(fc1_biases) +
118 |                 tf.nn.l2_loss(fc2_weights) + tf.nn.l2_loss(fc2_biases)+
119 |                 tf.nn.l2_loss(fc3_weights) + tf.nn.l2_loss(fc3_biases)+
120 |                 tf.nn.l2_loss(fc4_weights) + tf.nn.l2_loss(fc4_biases))
121 |         total_loss+=5e-4 * regularizers
122 |         train_op = tf.train.AdamOptimizer(learning_rate).minimize(total_loss)
123 | 
124 |         end_points['initial_state'] = initial_state
125 |         end_points['output'] = output
126 |         end_points['train_op'] = train_op
127 |         end_points['total_loss'] = total_loss
128 |         end_points['loss'] = loss
129 |         end_points['last_state'] = last_state
130 |         prediction = tf.nn.softmax(logits)
131 |         end_points['prediction'] = prediction
132 |     else:
133 |         prediction = tf.nn.softmax(logits)
134 | 
135 |         end_points['initial_state'] = initial_state
136 |         end_points['last_state'] = last_state
137 |         end_points['prediction'] = prediction
138 | 
139 |     return end_points
140 | 
141 | 


--------------------------------------------------------------------------------
/poems/poems.py:
--------------------------------------------------------------------------------
 1 | # -*- coding: utf-8 -*-
 2 | # file: poems.py
 3 | # author: JinTian
 4 | # time: 08/03/2017 7:39 PM
 5 | # Copyright 2017 JinTian. All Rights Reserved.
 6 | #
 7 | # Licensed under the Apache License, Version 2.0 (the "License");
 8 | # you may not use this file except in compliance with the License.
 9 | # You may obtain a copy of the License at
10 | #
11 | #     http://www.apache.org/licenses/LICENSE-2.0
12 | #
13 | # Unless required by applicable law or agreed to in writing, software
14 | # distributed under the License is distributed on an "AS IS" BASIS,
15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 | # See the License for the specific language governing permissions and
17 | # limitations under the License.
18 | # ------------------------------------------------------------------------
19 | import collections
20 | import os
21 | import sys
22 | import numpy as np
23 | 
24 | start_token = 'B'
25 | end_token = 'E'
26 | 
27 | 
28 | def process_poems(file_name):
29 |     # 诗集
30 |     poems = []
31 |     with open(file_name, "r", encoding='utf-8', ) as f:
32 |         for line in f.readlines():
33 |             try:
34 |                 title, content = line.strip().split(':')
35 |                 content = content.replace(' ', '')
36 |                 if '_' in content or '(' in content or '（' in content or '《' in content or '[' in content or \
37 |                         start_token in content or end_token in content:
38 |                     continue
39 |                 if len(content) < 5 or len(content) > 79:
40 |                     continue
41 |                 content = start_token + content + end_token
42 |                 poems.append(content)
43 |             except ValueError as e:
44 |                 pass
45 |     poems = sorted(poems, key=lambda l: len(line))
46 | 
47 |     all_words = []
48 |     for poem in poems:
49 |         all_words += [word for word in poem]
50 |     counter = collections.Counter(all_words)
51 |     count_pairs = sorted(counter.items(), key=lambda x: -x[1])
52 |     words, _ = zip(*count_pairs)
53 | 
54 |     words = words[:len(words)] + (' ',)
55 |     word_int_map = dict(zip(words, range(len(words))))
56 |     poems_vector = [list(map(lambda word: word_int_map.get(word, len(words)), poem)) for poem in poems]
57 | 
58 |     return poems_vector, word_int_map, words
59 | 
60 | 
61 | def generate_batch(batch_size, poems_vec, word_to_int):
62 |     n_chunk = len(poems_vec) // batch_size
63 |     x_batches = []
64 |     y_batches = []
65 |     for i in range(n_chunk):
66 |         start_index = i * batch_size
67 |         end_index = start_index + batch_size
68 | 
69 |         batches = poems_vec[start_index:end_index]
70 |         length = max(map(len, batches))
71 |         x_data = np.full((batch_size, length), word_to_int[' '], np.int32)
72 |         for row in range(batch_size):
73 |             x_data[row, :len(batches[row])] = batches[row]
74 |         y_data = np.copy(x_data)
75 |         y_data[:, :-1] = x_data[:, 1:]
76 |         """
77 |         x_data             y_data
78 |         [6,2,4,6,9]       [2,4,6,9,9]
79 |         [1,4,2,8,5]       [4,2,8,5,5]
80 |         """
81 |         x_batches.append(x_data)
82 |         y_batches.append(y_data)
83 |     return x_batches, y_batches
84 | 


--------------------------------------------------------------------------------
/poems/resnet.py:
--------------------------------------------------------------------------------
  1 | # Coder: Wenxin Xu
  2 | # Github: https://github.com/wenxinxu/resnet_in_tensorflow
  3 | # ==============================================================================
  4 | '''
  5 | This is the resnet structure
  6 | '''
  7 | import numpy as np
  8 | 
  9 | from hyper_parameters import *
 10 | 
 11 | BN_EPSILON = 0.001
 12 | 
 13 | 
 14 | def activation_summary(x):
 15 |     '''
 16 |     :param x: A Tensor
 17 |     :return: Add histogram summary and scalar summary of the sparsity of the tensor
 18 |     '''
 19 |     tensor_name = x.op.name
 20 |     tf.summary.histogram(tensor_name + '/activations', x)
 21 |     tf.summary.scalar(tensor_name + '/sparsity', tf.nn.zero_fraction(x))
 22 | 
 23 | 
 24 | def create_variables(name, shape, initializer=tf.contrib.layers.xavier_initializer(), is_fc_layer=False):
 25 |     '''
 26 |     :param name: A string. The name of the new variable
 27 |     :param shape: A list of dimensions
 28 |     :param initializer: User Xavier as default.
 29 |     :param is_fc_layer: Want to create fc layer variable? May use different weight_decay for fc
 30 |     layers.
 31 |     :return: The created variable
 32 |     '''
 33 | 
 34 |     ## TODO: to allow different weight decay to fully connected layer and conv layer
 35 |     if is_fc_layer is True:
 36 |         regularizer = tf.contrib.layers.l2_regularizer(scale=FLAGS.weight_decay)
 37 |     else:
 38 |         regularizer = tf.contrib.layers.l2_regularizer(scale=FLAGS.weight_decay)
 39 | 
 40 |     new_variables = tf.get_variable(name, shape=shape, initializer=initializer,
 41 |                                     regularizer=regularizer)
 42 |     return new_variables
 43 | 
 44 | 
 45 | def output_layer(input_layer, num_labels):
 46 |     '''
 47 |     :param input_layer: 2D tensor
 48 |     :param num_labels: int. How many output labels in total? (10 for cifar10 and 100 for cifar100)
 49 |     :return: output layer Y = WX + B
 50 |     '''
 51 |     input_dim = input_layer.get_shape().as_list()[-1]
 52 |     fc_w = create_variables(name='fc_weights', shape=[input_dim, num_labels], is_fc_layer=True,
 53 |                             initializer=tf.uniform_unit_scaling_initializer(factor=1.0))
 54 |     fc_b = create_variables(name='fc_bias', shape=[num_labels], initializer=tf.zeros_initializer())
 55 | 
 56 |     fc_h = tf.matmul(input_layer, fc_w) + fc_b
 57 |     return fc_h
 58 | 
 59 | 
 60 | def batch_normalization_layer(input_layer, dimension):
 61 |     '''
 62 |     Helper function to do batch normalziation
 63 |     :param input_layer: 4D tensor
 64 |     :param dimension: input_layer.get_shape().as_list()[-1]. The depth of the 4D tensor
 65 |     :return: the 4D tensor after being normalized
 66 |     '''
 67 |     mean, variance = tf.nn.moments(input_layer, axes=[0, 1, 2])
 68 |     beta = tf.get_variable('beta', dimension, tf.float32,
 69 |                            initializer=tf.constant_initializer(0.0, tf.float32))
 70 |     gamma = tf.get_variable('gamma', dimension, tf.float32,
 71 |                             initializer=tf.constant_initializer(1.0, tf.float32))
 72 |     bn_layer = tf.nn.batch_normalization(input_layer, mean, variance, beta, gamma, BN_EPSILON)
 73 | 
 74 |     return bn_layer
 75 | 
 76 | 
 77 | def conv_bn_relu_layer(input_layer, filter_shape, stride):
 78 |     '''
 79 |     A helper function to conv, batch normalize and relu the input tensor sequentially
 80 |     :param input_layer: 4D tensor
 81 |     :param filter_shape: list. [filter_height, filter_width, filter_depth, filter_number]
 82 |     :param stride: stride size for conv
 83 |     :return: 4D tensor. Y = Relu(batch_normalize(conv(X)))
 84 |     '''
 85 | 
 86 |     out_channel = filter_shape[-1]
 87 |     filter = create_variables(name='conv', shape=filter_shape)
 88 | 
 89 |     conv_layer = tf.nn.conv2d(input_layer, filter, strides=[1, stride, stride, 1], padding='SAME')
 90 |     bn_layer = batch_normalization_layer(conv_layer, out_channel)
 91 | 
 92 |     output = tf.nn.relu(bn_layer)
 93 |     return output
 94 | 
 95 | 
 96 | def bn_relu_conv_layer(input_layer, filter_shape, stride):
 97 |     '''
 98 |     A helper function to batch normalize, relu and conv the input layer sequentially
 99 |     :param input_layer: 4D tensor
100 |     :param filter_shape: list. [filter_height, filter_width, filter_depth, filter_number]
101 |     :param stride: stride size for conv
102 |     :return: 4D tensor. Y = conv(Relu(batch_normalize(X)))
103 |     '''
104 | 
105 |     in_channel = input_layer.get_shape().as_list()[-1]
106 | 
107 |     bn_layer = batch_normalization_layer(input_layer, in_channel)
108 |     relu_layer = tf.nn.relu(bn_layer)
109 | 
110 |     filter = create_variables(name='conv', shape=filter_shape)
111 |     conv_layer = tf.nn.conv2d(relu_layer, filter, strides=[1, stride, stride, 1], padding='SAME')
112 |     return conv_layer
113 | 
114 | 
115 | def residual_block(input_layer, output_channel, first_block=False):
116 |     '''
117 |     Defines a residual block in ResNet
118 |     :param input_layer: 4D tensor
119 |     :param output_channel: int. return_tensor.get_shape().as_list()[-1] = output_channel
120 |     :param first_block: if this is the first residual block of the whole network
121 |     :return: 4D tensor.
122 |     '''
123 |     input_channel = input_layer.get_shape().as_list()[-1]
124 | 
125 |     # When it's time to "shrink" the image size, we use stride = 2
126 |     if input_channel * 2 == output_channel:
127 |         increase_dim = True
128 |         stride = 2
129 |     elif input_channel == output_channel:
130 |         increase_dim = False
131 |         stride = 1
132 |     else:
133 |         raise ValueError('Output and input channel does not match in residual blocks!!!')
134 | 
135 |     # The first conv layer of the first residual block does not need to be normalized and relu-ed.
136 |     with tf.variable_scope('conv1_in_block'):
137 |         if first_block:
138 |             filter = create_variables(name='conv', shape=[3, 3, input_channel, output_channel])
139 |             conv1 = tf.nn.conv2d(input_layer, filter=filter, strides=[1, 1, 1, 1], padding='SAME')
140 |         else:
141 |             conv1 = bn_relu_conv_layer(input_layer, [3, 3, input_channel, output_channel], stride)
142 | 
143 |     with tf.variable_scope('conv2_in_block'):
144 |         conv2 = bn_relu_conv_layer(conv1, [3, 3, output_channel, output_channel], 1)
145 | 
146 |     # When the channels of input layer and conv2 does not match, we add zero pads to increase the
147 |     #  depth of input layers
148 |     if increase_dim is True:
149 |         pooled_input = tf.nn.avg_pool(input_layer, ksize=[1, 1, 1, 1],
150 |                                       strides=[1, 2, 2, 1], padding='VALID')
151 |         padded_input = tf.pad(pooled_input, [[0, 0], [0, 0], [0, 0], [input_channel // 2,
152 |                                                                       input_channel // 2]])
153 |     else:
154 |         padded_input = input_layer
155 | 
156 |     output = conv2 + padded_input
157 |     return output
158 | 
159 | 
160 | def inference(input_tensor_batch, n, reuse,output_num=128):
161 |     '''
162 |     The main function that defines the ResNet. total layers = 1 + 2n + 2n + 2n +1 = 6n + 2
163 |     :param input_tensor_batch: 4D tensor
164 |     :param n: num_residual_blocks
165 |     :param reuse: To build train graph, reuse=False. To build validation graph and share weights
166 |     with train graph, resue=True
167 |     :return: last layer in the network. Not softmax-ed
168 |     '''
169 | 
170 |     layers = []
171 |     with tf.variable_scope('conv0', reuse=reuse):
172 |         conv0 = conv_bn_relu_layer(input_tensor_batch, [1, 1, 1, 16], 1)
173 |         activation_summary(conv0)
174 |         layers.append(conv0)
175 | 
176 |     for i in range(n):
177 |         with tf.variable_scope('conv1_%d' % i, reuse=reuse):
178 |             if i == 0:
179 |                 conv1 = residual_block(layers[-1], 16, first_block=True)
180 |             else:
181 |                 conv1 = residual_block(layers[-1], 16)
182 |             activation_summary(conv1)
183 |             layers.append(conv1)
184 | 
185 |     for i in range(n):
186 |         with tf.variable_scope('conv2_%d' % i, reuse=reuse):
187 |             conv2 = residual_block(layers[-1], 32)
188 |             activation_summary(conv2)
189 |             layers.append(conv2)
190 | 
191 |     for i in range(n):
192 |         with tf.variable_scope('conv3_%d' % i, reuse=reuse):
193 |             conv3 = residual_block(layers[-1], 64)
194 |             layers.append(conv3)
195 |         # assert conv3.get_shape().as_list()[1:] == [8, 8, 64]
196 | 
197 |     with tf.variable_scope('fc', reuse=reuse):
198 |         in_channel = layers[-1].get_shape().as_list()[-1]
199 |         bn_layer = batch_normalization_layer(layers[-1], in_channel)
200 |         relu_layer = tf.nn.relu(bn_layer)
201 |         global_pool = tf.reduce_mean(relu_layer, [1, 2])
202 | 
203 |         assert global_pool.get_shape().as_list()[-1:] == [64]
204 |         output = output_layer(global_pool, output_num)
205 |         layers.append(output)
206 | 
207 |     return layers[-1]
208 | 
209 | 
210 | def test_graph(train_dir='logs'):
211 |     '''
212 |     Run this function to look at the graph structure on tensorboard. A fast way!
213 |     :param train_dir:
214 |     '''
215 |     input_tensor = tf.constant(np.ones([128, 32, 32, 3]), dtype=tf.float32)
216 |     result = inference(input_tensor, 2, reuse=False)
217 |     init = tf.initialize_all_variables()
218 |     sess = tf.Session()
219 |     sess.run(init)
220 |     summary_writer = tf.summary.FileWriter(train_dir, sess.graph)


--------------------------------------------------------------------------------
/ssq.py:
--------------------------------------------------------------------------------
  1 | # -*- coding: utf-8 -*-
  2 | # file: main.py
  3 | # author: JinTian
  4 | # time: 11/03/2017 9:53 AM
  5 | # Copyright 2017 JinTian. All Rights Reserved.
  6 | #
  7 | # Licensed under the Apache License, Version 2.0 (the "License");
  8 | # you may not use this file except in compliance with the License.
  9 | # You may obtain a copy of the License at
 10 | #
 11 | #     http://www.apache.org/licenses/LICENSE-2.0
 12 | #
 13 | # Unless required by applicable law or agreed to in writing, software
 14 | # distributed under the License is distributed on an "AS IS" BASIS,
 15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 16 | # See the License for the specific language governing permissions and
 17 | # limitations under the License.
 18 | # ------------------------------------------------------------------------
 19 | import os
 20 | import numpy as np
 21 | import tensorflow as tf
 22 | from poems.model import rnn_model
 23 | from poems.poems import process_poems, generate_batch
 24 | from ssq_data import *
 25 | tf.app.flags.DEFINE_integer('batch_size', -1, 'batch size.')
 26 | tf.app.flags.DEFINE_float('learning_rate', 0.0001, 'learning rate.')
 27 | tf.app.flags.DEFINE_string('model_dir', os.path.abspath('./model'), 'model save path.')
 28 | tf.app.flags.DEFINE_string('file_path', os.path.abspath('./data/poems.txt'), 'file name of poems.')
 29 | tf.app.flags.DEFINE_string('model_prefix', 'poems', 'model save prefix.')
 30 | tf.app.flags.DEFINE_integer('epochs', 150000, 'train how many epochs.')
 31 | 
 32 | FLAGS = tf.app.flags.FLAGS
 33 | 
 34 | 
 35 | def run_training():
 36 |     # if not os.path.exists(FLAGS.model_dir):
 37 |     #     os.makedirs(FLAGS.model_dir)
 38 |     #
 39 |     # poems_vector, word_to_int, vocabularies = process_poems(FLAGS.file_path)
 40 |     # batches_inputs, batches_outputs = generate_batch(FLAGS.batch_size, poems_vector, word_to_int)
 41 |     ssqdata=get_exl_data(random_order=True)
 42 |     batches_inputs=ssqdata[0:(len(ssqdata)-1)]
 43 |     batches_outputs = ssqdata[1:(len(ssqdata))]
 44 |     FLAGS.batch_size=len(ssqdata)-1
 45 |     # data=batches_outputs[1:7]
 46 |     # print(len(data))
 47 |     del ssqdata
 48 |     input_data = tf.placeholder(tf.int32, [FLAGS.batch_size, None])
 49 |     # print(tf.shape(input_data))
 50 |     output_targets = tf.placeholder(tf.int32, [FLAGS.batch_size, None])
 51 |     end_points = rnn_model(model='lstm', input_data=input_data, output_data=output_targets, vocab_size=33+16,
 52 |                            output_num=7,input_num=7,
 53 |                            rnn_size=128, num_layers=7, batch_size=FLAGS.batch_size, learning_rate=FLAGS.learning_rate)
 54 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=output_targets, vocab_size=len(
 55 |     #     vocabularies), rnn_size=128, num_layers=2, batch_size=64, learning_rate=FLAGS.learning_rate)
 56 | 
 57 |     saver = tf.train.Saver(tf.global_variables())
 58 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
 59 |     with tf.Session() as sess:
 60 |         # sess = tf_debug.LocalCLIDebugWrapperSession(sess=sess)
 61 |         # sess.add_tensor_filter("has_inf_or_nan", tf_debug.has_inf_or_nan)
 62 |         sess.run(init_op)
 63 | 
 64 |         start_epoch = 0
 65 |         # saver.restore(sess, "F:/tensorflow_poems-master/model/poems-100000")
 66 |         checkpoint = tf.train.latest_checkpoint(FLAGS.model_dir)
 67 |         if checkpoint:
 68 |             saver.restore(sess, checkpoint)
 69 |             print("## restore from the checkpoint {0}".format(checkpoint))
 70 |             start_epoch += int(checkpoint.split('-')[-1])
 71 |         print('## start training...')
 72 |         try:
 73 |             for epoch in range(start_epoch, FLAGS.epochs):
 74 |                 n = 0
 75 |                 # n_chunk = len(poems_vector) // FLAGS.batch_size
 76 |                 # n_chunk = len(batches_inputs) // FLAGS.batch_size
 77 |                 n_chunk=math.ceil(len(batches_inputs) / FLAGS.batch_size)
 78 |                 for batch in range(n_chunk):
 79 |                     left=(batch+1)*FLAGS.batch_size-len(batches_inputs)
 80 |                     if left<0:
 81 |                         inputdata=batches_inputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 82 |                         outputdata=batches_outputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 83 |                     else:
 84 |                         # temp=batches_inputs[batch*FLAGS.batch_size:len(batches_inputs) ]
 85 |                         # temp.extend(batches_inputs[0:left])
 86 |                         inputdata=batches_inputs[len(batches_inputs)-FLAGS.batch_size:len(batches_inputs)]
 87 |                         # temp=batches_outputs[batch*FLAGS.batch_size:len(batches_inputs) ]
 88 |                         # temp.extend(batches_outputs[0:left])
 89 |                         outputdata=batches_outputs[len(batches_outputs)-FLAGS.batch_size:len(batches_outputs)]
 90 |                     # print(len(inputdata))
 91 |                     loss, _, _ = sess.run([
 92 |                         end_points['total_loss'],
 93 |                         end_points['last_state'],
 94 |                         end_points['train_op']
 95 |                     ], feed_dict={input_data: inputdata, output_targets: outputdata})
 96 |                     # # ], feed_dict={input_data: batches_inputs[n], output_targets: batches_outputs[n]})
 97 |                     n += 1
 98 |                     print('Epoch: %d, batch: %d, training loss: %.6f' % (epoch, batch, loss))
 99 |                 if epoch % 5000 == 0:
100 |                     saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
101 |         except KeyboardInterrupt:
102 |             print('## Interrupt manually, try saving checkpoint for now...')
103 |         finally:
104 |             saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
105 |             print('## Last epoch were saved, next time will start from epoch {}.'.format(epoch))
106 | 
107 | 
108 | def main(_):
109 |     run_training()
110 | 
111 | 
112 | if __name__ == '__main__':
113 |     tf.app.run()
114 | 


--------------------------------------------------------------------------------
/ssq.xls:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Liang-yc/ssq/4c69745d1fb731b4a39cc571af58a31368439789/ssq.xls


--------------------------------------------------------------------------------
/ssq4all.py:
--------------------------------------------------------------------------------
  1 | # -*- coding: utf-8 -*-
  2 | # file: main.py
  3 | # author: JinTian
  4 | # time: 11/03/2017 9:53 AM
  5 | # Copyright 2017 JinTian. All Rights Reserved.
  6 | #
  7 | # Licensed under the Apache License, Version 2.0 (the "License");
  8 | # you may not use this file except in compliance with the License.
  9 | # You may obtain a copy of the License at
 10 | #
 11 | #     http://www.apache.org/licenses/LICENSE-2.0
 12 | #
 13 | # Unless required by applicable law or agreed to in writing, software
 14 | # distributed under the License is distributed on an "AS IS" BASIS,
 15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 16 | # See the License for the specific language governing permissions and
 17 | # limitations under the License.
 18 | # ------------------------------------------------------------------------
 19 | import os
 20 | import numpy as np
 21 | import tensorflow as tf
 22 | from poems.model import rnn_model
 23 | from poems.resnet import *
 24 | from poems.poems import process_poems, generate_batch
 25 | from ssq_data import *
 26 | # for Windows10：OSError: raw write() returned invalid length 96 (should have been between 0 and 48)
 27 | # import win_unicode_console
 28 | # win_unicode_console.enable()
 29 | tf.app.flags.DEFINE_integer('batch_size', 2214, 'batch size.')
 30 | tf.app.flags.DEFINE_float('learning_rate', 0.0001, 'learning rate.')
 31 | tf.app.flags.DEFINE_string('model_dir', os.path.abspath('./model4all'), 'model save path.')
 32 | tf.app.flags.DEFINE_string('file_path', os.path.abspath('./data/poems.txt'), 'file name of poems.')
 33 | tf.app.flags.DEFINE_string('model_prefix', 'poems', 'model save prefix.')
 34 | tf.app.flags.DEFINE_integer('epochs', 500000, 'train how many epochs.')
 35 | 
 36 | FLAGS = tf.app.flags.FLAGS
 37 | 
 38 | 
 39 | def run_training():
 40 |     if not os.path.exists(FLAGS.model_dir):
 41 |         os.makedirs(FLAGS.model_dir)
 42 |     #
 43 |     # poems_vector, word_to_int, vocabularies = process_poems(FLAGS.file_path)
 44 |     # batches_inputs, batches_outputs = generate_batch(FLAGS.batch_size, poems_vector, word_to_int)
 45 |     ssqdata=get_exl_data(random_order=False,use_resnet=True)
 46 |     # print(ssqdata[len(ssqdata)-1])
 47 |     batches_inputs=ssqdata[0:(len(ssqdata)-1)]
 48 |     ssqdata=get_exl_data(random_order=False,use_resnet=False)
 49 |     batches_outputs = ssqdata[1:(len(ssqdata))]
 50 |     FLAGS.batch_size=len(batches_inputs)
 51 |     # print(np.shape(batches_outputs))
 52 |     # data=batches_outputs[1:7]
 53 |     # print(len(data))
 54 |     del ssqdata
 55 |     input_data = tf.placeholder(tf.float32, [FLAGS.batch_size, 1,7,1])
 56 |     logits = inference(input_data, 2, reuse=False,output_num=128)
 57 | 
 58 |     # print(tf.shape(input_data))
 59 |     output_targets = tf.placeholder(tf.int32, [FLAGS.batch_size, None])
 60 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=33+16,output_num=7,
 61 |                            rnn_size=128, num_layers=7, batch_size=FLAGS.batch_size, learning_rate=FLAGS.learning_rate)
 62 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=output_targets, vocab_size=len(
 63 |     #     vocabularies), rnn_size=128, num_layers=2, batch_size=64, learning_rate=FLAGS.learning_rate)
 64 | 
 65 |     saver = tf.train.Saver(tf.global_variables())
 66 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
 67 |     with tf.Session() as sess:
 68 |         # sess = tf_debug.LocalCLIDebugWrapperSession(sess=sess)
 69 |         # sess.add_tensor_filter("has_inf_or_nan", tf_debug.has_inf_or_nan)
 70 |         sess.run(init_op)
 71 | 
 72 |         start_epoch = 0
 73 |         checkpoint = tf.train.latest_checkpoint(FLAGS.model_dir)
 74 |         if checkpoint:
 75 |             saver.restore(sess, checkpoint)
 76 |             print("## restore from the checkpoint {0}".format(checkpoint))
 77 |             start_epoch += int(checkpoint.split('-')[-1])
 78 |         print('## start training...')
 79 |         try:
 80 |             for epoch in range(start_epoch, FLAGS.epochs):
 81 |                 n = 0
 82 |                 # n_chunk = len(poems_vector) // FLAGS.batch_size
 83 |                 # n_chunk = len(batches_inputs) // FLAGS.batch_size
 84 |                 n_chunk=math.ceil(len(batches_inputs) / FLAGS.batch_size)
 85 |                 for batch in range(n_chunk):
 86 |                     left=(batch+1)*FLAGS.batch_size-len(batches_inputs)
 87 |                     if left<0:
 88 |                         inputdata=batches_inputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 89 |                         outputdata=batches_outputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 90 |                     else:
 91 |                         # temp=batches_inputs[batch*FLAGS.batch_size:len(batches_inputs) ]
 92 |                         # temp.extend(batches_inputs[0:left])
 93 |                         inputdata=batches_inputs[len(batches_inputs)-FLAGS.batch_size:len(batches_inputs)]
 94 |                         # temp=batches_outputs[batch*FLAGS.batch_size:len(batches_inputs) ]
 95 |                         # temp.extend(batches_outputs[0:left])
 96 |                         outputdata=batches_outputs[len(batches_outputs)-FLAGS.batch_size:len(batches_outputs)]
 97 |                     # print(len(inputdata))
 98 |                     loss, _, _ = sess.run([
 99 |                         end_points['total_loss'],
100 |                         end_points['last_state'],
101 |                         end_points['train_op']
102 |                     ], feed_dict={input_data: inputdata, output_targets: outputdata})
103 |                     # ], feed_dict={input_data: batches_inputs, output_targets: batches_outputs})
104 |                     n += 1
105 |                     print('Epoch: %d, batch: %d, training loss: %.6f' % (epoch, batch, loss))
106 |                 if epoch % 50000 == 0:
107 |                     saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
108 |         except KeyboardInterrupt:
109 |             print('## Interrupt manually, try saving checkpoint for now...')
110 |         finally:
111 |             saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
112 |             print('## Last epoch were saved, next time will start from epoch {}.'.format(epoch))
113 | 
114 | 
115 | def main(_):
116 |     run_training()
117 | 
118 | 
119 | if __name__ == '__main__':
120 |     tf.app.run()
121 | 


--------------------------------------------------------------------------------
/ssq4all_test.py:
--------------------------------------------------------------------------------
  1 | # -*- coding: utf-8 -*-
  2 | # file: main.py
  3 | # author: JinTian
  4 | # time: 11/03/2017 9:53 AM
  5 | # Copyright 2017 JinTian. All Rights Reserved.
  6 | #
  7 | # Licensed under the Apache License, Version 2.0 (the "License");
  8 | # you may not use this file except in compliance with the License.
  9 | # You may obtain a copy of the License at
 10 | #
 11 | #     http://www.apache.org/licenses/LICENSE-2.0
 12 | #
 13 | # Unless required by applicable law or agreed to in writing, software
 14 | # distributed under the License is distributed on an "AS IS" BASIS,
 15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 16 | # See the License for the specific language governing permissions and
 17 | # limitations under the License.
 18 | # ------------------------------------------------------------------------
 19 | import tensorflow as tf
 20 | from poems.model import rnn_model
 21 | from poems.resnet import *
 22 | from poems.poems import process_poems
 23 | import numpy as np
 24 | from ssq_data import *
 25 | start_token = 'B'
 26 | end_token = 'E'
 27 | model_dir = './model4all/'
 28 | corpus_file = './data/poems.txt'
 29 | 
 30 | 
 31 | def to_word(predict, vocabs):
 32 |     t = np.cumsum(predict)
 33 |     s = np.sum(predict)
 34 |     sample = int(np.searchsorted(t, np.random.rand(1) * s))
 35 |     if sample > len(vocabs):
 36 |         sample = len(vocabs) - 1
 37 |     return vocabs[sample]
 38 | 
 39 | 
 40 | def gen_poem():
 41 |     batch_size = 1
 42 |     print('## loading model from %s' % model_dir)
 43 |     input_data = tf.placeholder(tf.float32, [1, 1,7,1])
 44 |     logits = inference(input_data, 2, reuse=False,output_num=128)
 45 | 
 46 |     # print(tf.shape(input_data))
 47 |     output_targets = tf.placeholder(tf.int32, [1, None])
 48 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=33+16,
 49 |                            output_num=7,
 50 |                            rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.01)
 51 | 
 52 |     # input_data = tf.placeholder(tf.int32, [batch_size, None])
 53 |     #
 54 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=None, vocab_size=33,
 55 |     #                        rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.01)
 56 | 
 57 |     saver = tf.train.Saver(tf.global_variables())
 58 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
 59 |     with tf.Session() as sess:
 60 |         sess.run(init_op)
 61 | 
 62 |         checkpoint = tf.train.latest_checkpoint('./model4all/')
 63 |         saver.restore(sess, checkpoint)
 64 |         # saver.restore(sess, "F:/tensorflow_poems-master/model4all/poems-401713")
 65 |         ssqdata = get_exl_data(random_order=True,use_resnet=True)
 66 |         # x = np.array([list(map(word_int_map.get, start_token))])
 67 |         x=[ssqdata[len(ssqdata)-1]]
 68 |         print("input: %s"%(x+np.asarray([[[[1],[1],[1],[1],[1],[1],[-32]]]])))
 69 |         [predict, last_state] = sess.run([end_points['prediction'], end_points['last_state']],
 70 |                                          feed_dict={input_data: x})
 71 |         poem_=np.argmax(np.array(predict),axis=1)
 72 |         sorted_result = np.argsort(np.array(predict), axis=1)
 73 |         results=poem_+np.asarray([1,1,1,1,1,1,-32])
 74 |         print(sorted_result)
 75 |         print("output: %s"%results)
 76 |         poem_=np.argmin(np.array(predict),axis=1)
 77 |         results=poem_+np.asarray([1,1,1,1,1,1,-32])
 78 |         print("min output:%s"%results)
 79 |         return poem_
 80 | 
 81 | def gen_blue():
 82 |     batch_size = 1
 83 |     print('## loading model from %s' % model_dir)
 84 |     input_data = tf.placeholder(tf.float32, [1, 1,7,1])
 85 |     logits = inference(input_data, 10, reuse=False,output_num=128)
 86 | 
 87 |     # print(tf.shape(input_data))
 88 |     output_targets = tf.placeholder(tf.int32, [1, None])
 89 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=33,
 90 |                            output_num=1,
 91 |                            rnn_size=128, num_layers=3, batch_size=1, learning_rate=0.01)
 92 | 
 93 |     # input_data = tf.placeholder(tf.int32, [batch_size, None])
 94 |     #
 95 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=None, vocab_size=33,
 96 |     #                        rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.01)
 97 | 
 98 |     saver = tf.train.Saver(tf.global_variables())
 99 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
100 |     with tf.Session() as sess:
101 |         sess.run(init_op)
102 | 
103 |         checkpoint = tf.train.latest_checkpoint('./model4all/')
104 |         saver.restore(sess, checkpoint)
105 |         # saver.restore(sess, "E:/workplace/tensorflow_poems-master/model4blue/poems-84201")
106 |         ssqdata = get_exl_data(random_order=True,use_resnet=True)
107 |         # x = np.array([list(map(word_int_map.get, start_token))])
108 |         x=[ssqdata[len(ssqdata)-1]]
109 | 
110 |         print("input: %s"%(x))
111 |         [predict, last_state] = sess.run([end_points['prediction'], end_points['last_state']],
112 |                                          feed_dict={input_data: x})
113 |         poem_=np.argmax(np.array(predict),axis=1)
114 |         sorted_result=np.argsort(np.array(predict),axis=1)
115 |         results=poem_+np.ones(1)
116 |         print(results)
117 |         print(sorted_result)
118 |         return poem_
119 | 
120 | 
121 | if __name__ == '__main__':
122 |     # begin_char = input('## please input the first character:')
123 |     # poem=gen_blue()#13
124 |     poem = gen_poem()
125 |     # pretty_print_poem(poem_=poem)


--------------------------------------------------------------------------------
/ssq4all_test_v4.py:
--------------------------------------------------------------------------------
 1 | # -*- coding: utf-8 -*-
 2 | # file: main.py
 3 | # author: JinTian
 4 | # time: 11/03/2017 9:53 AM
 5 | # Copyright 2017 JinTian. All Rights Reserved.
 6 | #
 7 | # Licensed under the Apache License, Version 2.0 (the "License");
 8 | # you may not use this file except in compliance with the License.
 9 | # You may obtain a copy of the License at
10 | #
11 | #     http://www.apache.org/licenses/LICENSE-2.0
12 | #
13 | # Unless required by applicable law or agreed to in writing, software
14 | # distributed under the License is distributed on an "AS IS" BASIS,
15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 | # See the License for the specific language governing permissions and
17 | # limitations under the License.
18 | # ------------------------------------------------------------------------
19 | import tensorflow as tf
20 | from poems.model import rnn_model
21 | from poems.resnet import *
22 | from poems.poems import process_poems
23 | import numpy as np
24 | from ssq_data import *
25 | start_token = 'B'
26 | end_token = 'E'
27 | model_dir = './model4all_v2/'
28 | corpus_file = './data/poems.txt'
29 | 
30 | 
31 | def to_word(predict, vocabs):
32 |     t = np.cumsum(predict)
33 |     s = np.sum(predict)
34 |     sample = int(np.searchsorted(t, np.random.rand(1) * s))
35 |     if sample > len(vocabs):
36 |         sample = len(vocabs) - 1
37 |     return vocabs[sample]
38 | 
39 | 
40 | def gen_poem():
41 |     batch_size = 1
42 |     print('## loading model from %s' % model_dir)
43 |     input_data = tf.placeholder(tf.float32, [1, 10,7+8,1])
44 |     logits = inference(input_data, 1, reuse=False,output_num=128)
45 | 
46 |     # print(tf.shape(input_data))
47 |     output_targets = tf.placeholder(tf.int32, [1, None])
48 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=33+16,output_num=7,
49 |                            rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.001)
50 | 
51 |     # input_data = tf.placeholder(tf.int32, [batch_size, None])
52 |     #
53 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=None, vocab_size=33,
54 |     #                        rnn_size=128, num_layers=7, batch_size=1, learning_rate=0.01)
55 | 
56 |     saver = tf.train.Saver(tf.global_variables())
57 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
58 |     with tf.Session() as sess:
59 |         sess.run(init_op)
60 | 
61 |         checkpoint = tf.train.latest_checkpoint('./model4all_v4/')
62 |         saver.restore(sess, checkpoint)
63 |         # saver.restore(sess, "E:/workplace/tensorflow_poems-master/model4all/poems-208368")
64 |         # ssqdata = get_exl_data(random_order=True,use_resnet=True)
65 |         # ssqdata = get_exl_data_v3(random_order=True, use_resnet=True)
66 |         ssqdata = get_exl_data_by_period(random_order=True, use_resnet=True, times=10)
67 |         x=[ssqdata[len(ssqdata)-1]]
68 |         print("input: %s"%(x+np.asarray([[[[1],[1],[1],[1],[1],[1],[-32],[0],[0],[0],[0],[0],[0],[0],[0]]]])))
69 |         [predict, last_state] = sess.run([end_points['prediction'], end_points['last_state']],
70 |                                          feed_dict={input_data: x})
71 |         poem_=np.argmax(np.array(predict),axis=1)
72 |         sorted_result = np.argsort(np.array(predict), axis=1)
73 |         results=poem_+np.asarray([1,1,1,1,1,1,-32])
74 |         print(sorted_result)
75 |         print("output: %s"%results)
76 |         return poem_
77 | 
78 | 
79 | 
80 | 
81 | if __name__ == '__main__':
82 |     # begin_char = input('## please input the first character:')
83 |     # poem=gen_blue()#13
84 |     poem = gen_poem()
85 |     # pretty_print_poem(poem_=poem)
86 | 


--------------------------------------------------------------------------------
/ssq4all_v4.py:
--------------------------------------------------------------------------------
  1 | # *_*coding:utf-8 *_*
  2 | 
  3 | import os
  4 | import sys
  5 | sys.path.append(os.path.abspath(os.path.join(os.getcwd(),"..")))
  6 | import numpy as np
  7 | import tensorflow as tf
  8 | from poems.model import rnn_model
  9 | from poems.resnet import *
 10 | from poems.poems import process_poems, generate_batch
 11 | from ssq_data import *
 12 | # for Windows10：OSError: raw write() returned invalid length 96 (should have been between 0 and 48)
 13 | import win_unicode_console
 14 | win_unicode_console.enable()
 15 | tf.app.flags.DEFINE_integer('batch_size', 2214, 'batch size.')
 16 | tf.app.flags.DEFINE_float('learning_rate', 0.0001, 'learning rate.')
 17 | tf.app.flags.DEFINE_string('model_dir', os.path.abspath('./model4all_v4'), 'model save path.')
 18 | tf.app.flags.DEFINE_string('file_path', os.path.abspath('./data/poems.txt'), 'file name of poems.')
 19 | tf.app.flags.DEFINE_string('model_prefix', 'poems', 'model save prefix.')
 20 | tf.app.flags.DEFINE_integer('epochs', 2000001, 'train how many epochs.')
 21 | tf.app.flags.DEFINE_integer('times', 10, 'train how many epochs.')
 22 | FLAGS = tf.app.flags.FLAGS
 23 | 
 24 | 
 25 | def run_training():
 26 |     if not os.path.exists(FLAGS.model_dir):
 27 |         os.makedirs(FLAGS.model_dir)
 28 |     #
 29 |     # poems_vector, word_to_int, vocabularies = process_poems(FLAGS.file_path)
 30 |     # batches_inputs, batches_outputs = generate_batch(FLAGS.batch_size, poems_vector, word_to_int)
 31 |     # ssqdata=get_exl_data(random_order=False,use_resnet=True)
 32 |     # # print(ssqdata[len(ssqdata)-1])
 33 |     # batches_inputs=ssqdata[0:(len(ssqdata)-1)]
 34 |     ssqdata=get_exl_data(random_order=False,use_resnet=False)
 35 |     # ssqdata=get_dlt_data(random_order=False,use_resnet=True)
 36 |     # print(ssqdata[len(ssqdata)-1])
 37 |     batches_inputs=ssqdata[0:(len(ssqdata)-1)]
 38 |     # ssqdata=get_dlt_data(random_order=False,use_resnet=False)
 39 |     batches_outputs = ssqdata[1:(len(ssqdata))]
 40 |     ssqdata=get_exl_data_by_period(random_order=False,use_resnet=True,times=FLAGS.times)
 41 |     batches_inputs=ssqdata[0:(len(ssqdata)-1)]
 42 |     FLAGS.batch_size=len(batches_inputs)
 43 |     # print(np.shape(batches_outputs))
 44 |     # data=batches_outputs[1:7]
 45 |     # print(len(data))
 46 |     del ssqdata
 47 |     input_data = tf.placeholder(tf.float32, [FLAGS.batch_size, FLAGS.times,7+8,1])
 48 |     logits = inference(input_data, 1, reuse=False,output_num=128)
 49 | 
 50 |     # print(tf.shape(input_data))
 51 |     output_targets = tf.placeholder(tf.int32, [FLAGS.batch_size, None])
 52 |     end_points = rnn_model(model='lstm', input_data=logits, output_data=output_targets, vocab_size=33+16,output_num=7,
 53 |                            rnn_size=128, num_layers=7, batch_size=FLAGS.batch_size, learning_rate=FLAGS.learning_rate)
 54 |     # end_points = rnn_model(model='lstm', input_data=input_data, output_data=output_targets, vocab_size=len(
 55 |     #     vocabularies), rnn_size=128, num_layers=2, batch_size=64, learning_rate=FLAGS.learning_rate)
 56 | 
 57 |     saver = tf.train.Saver(tf.global_variables())
 58 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
 59 |     with tf.Session() as sess:
 60 |         # sess = tf_debug.LocalCLIDebugWrapperSession(sess=sess)
 61 |         # sess.add_tensor_filter("has_inf_or_nan", tf_debug.has_inf_or_nan)
 62 |         sess.run(init_op)
 63 | 
 64 |         start_epoch = 0
 65 |         # saver.restore(sess, "E:/workplace/tensorflow_poems-master/model4all/poems-208368")
 66 |         checkpoint = tf.train.latest_checkpoint(FLAGS.model_dir)
 67 |         if checkpoint:
 68 |             saver.restore(sess, checkpoint)
 69 |             print("## restore from the checkpoint {0}".format(checkpoint))
 70 |             start_epoch += int(checkpoint.split('-')[-1])
 71 |         print('## start training...')
 72 |         try:
 73 |             # for epoch in range(start_epoch, FLAGS.epochs):
 74 |             epoch = start_epoch
 75 |             FLAGS.epochs=epoch+100000
 76 |             print('till',FLAGS.epochs)
 77 |             # for epoch in range(start_epoch, FLAGS.epochs):
 78 |             while(epoch<FLAGS.epochs):
 79 |                 epoch = epoch + 1
 80 |                 n = 0
 81 |                 # n_chunk = len(poems_vector) // FLAGS.batch_size
 82 |                 # n_chunk = len(batches_inputs) // FLAGS.batch_size
 83 |                 n_chunk=math.ceil(len(batches_inputs) / FLAGS.batch_size)
 84 |                 for batch in range(n_chunk):
 85 |                     left=(batch+1)*FLAGS.batch_size-len(batches_inputs)
 86 |                     if left<0:
 87 |                         inputdata=batches_inputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 88 |                         outputdata=batches_outputs[(batch*FLAGS.batch_size):((batch+1)*FLAGS.batch_size)]
 89 |                     else:
 90 |                         # temp=batches_inputs[batch*FLAGS.batch_size:len(batches_inputs) ]
 91 |                         # temp.extend(batches_inputs[0:left])
 92 |                         inputdata=batches_inputs[len(batches_inputs)-FLAGS.batch_size:len(batches_inputs)]
 93 |                         # temp=batches_outputs[batch*FLAGS.batch_size:len(batches_inputs) ]
 94 |                         # temp.extend(batches_outputs[0:left])
 95 |                         outputdata=batches_outputs[len(batches_outputs)-FLAGS.batch_size:len(batches_outputs)]
 96 |                     # print(len(inputdata))
 97 |                     loss, _, _ = sess.run([
 98 |                         end_points['total_loss'],
 99 |                         end_points['last_state'],
100 |                         end_points['train_op']
101 |                     ], feed_dict={input_data: inputdata, output_targets: outputdata})
102 |                     # ], feed_dict={input_data: batches_inputs, output_targets: batches_outputs})
103 |                     n += 1
104 |                 if epoch % 100 == 0:
105 |                     print('Epoch: %d, batch: %d, training loss: %.6f' % (epoch, batch, loss))
106 |                 if epoch % 50000 == 0:
107 |                     saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
108 |         except KeyboardInterrupt:
109 |             print('## Interrupt manually, try saving checkpoint for now...')
110 |         finally:
111 |             saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
112 |             print('## Last epoch were saved, next time will start from epoch {}.'.format(epoch))
113 |         # saver.save(sess, os.path.join(FLAGS.model_dir, FLAGS.model_prefix), global_step=epoch)
114 | 
115 | 
116 | def main(_):
117 |     run_training()
118 | 
119 | 
120 | if __name__ == '__main__':
121 |     tf.app.run()
122 | 


--------------------------------------------------------------------------------
/ssq_data.py:
--------------------------------------------------------------------------------
  1 | 
  2 | import tensorflow as tf
  3 | # import matplotlib.pyplot as plt
  4 | from collections import OrderedDict
  5 | import math
  6 | from pyexcel_xls import get_data
  7 | from pyexcel_xls import save_data
  8 | import numpy as np
  9 | import cv2
 10 | from PIL import Image
 11 | import os
 12 | def get_exl_data_v2(exl_path='./ssq.xls',random_order=False,use_resnet=False):
 13 |     ssq_data=[]
 14 |     xls_data = get_data(exl_path)
 15 |     # print(type(xls_data))
 16 |     if use_resnet:
 17 |         if random_order:
 18 |             for i in range(2, len(xls_data['data'])):
 19 |                 # print(type(xls_data['data'][i][2]))
 20 |                 # aaaa=xls_data['data'][i][2]
 21 |                 if xls_data['data'][i] == []:
 22 |                     break
 23 |                 temp = np.asarray([[
 24 |                     [xls_data['data'][i][9]-1],
 25 |                      [xls_data['data'][i][10]-1],
 26 |                      [xls_data['data'][i][11]-1],
 27 |                      [xls_data['data'][i][12]-1],
 28 |                      [xls_data['data'][i][13]-1],
 29 |                      [xls_data['data'][i][14]-1],
 30 |                      [xls_data['data'][i][8]+32],
 31 |                     [xls_data['data'][i][17]/10000000],
 32 |                     [xls_data['data'][i][19]/10000000],
 33 |                     [xls_data['data'][i][21]/10000000],
 34 |                     [xls_data['data'][i][23] / 10000000],
 35 |                     [xls_data['data'][i][25] / 10000000],
 36 |                     [xls_data['data'][i][27] / 10000000]
 37 |                 ]])
 38 |                 # print(temp.shape)
 39 |                 #
 40 |                 ssq_data.append(temp)
 41 |         else:
 42 |             for i in range(2, len(xls_data['data'])):
 43 |                 # print(type(xls_data['data'][i][2]))
 44 |                 # aaaa=xls_data['data'][i][2]
 45 |                 if xls_data['data'][i] == []:
 46 |                     break
 47 |                 temp = np.asarray([[
 48 |                     [xls_data['data'][i][2]-1],
 49 |                     [xls_data['data'][i][3]-1],
 50 |                     [xls_data['data'][i][4]-1],
 51 |                     [xls_data['data'][i][5]-1],
 52 |                     [xls_data['data'][i][6]-1],
 53 |                     [xls_data['data'][i][7]-1],
 54 |                     [xls_data['data'][i][8] + 33],
 55 |                     [xls_data['data'][i][17]/10000000],
 56 |                     [xls_data['data'][i][19]/10000000],
 57 |                     [xls_data['data'][i][21]/10000000],
 58 |                     [xls_data['data'][i][23] / 10000000],
 59 |                     [xls_data['data'][i][25] / 10000000],
 60 |                     [xls_data['data'][i][27] / 10000000]]])
 61 |                 ssq_data.append(temp)
 62 |         return ssq_data
 63 | 
 64 |     if random_order:
 65 |         for i in range(2, len(xls_data['data'])):
 66 |             # print(type(xls_data['data'][i][2]))
 67 |             # aaaa=xls_data['data'][i][2]
 68 |             if xls_data['data'][i] == []:
 69 |                 break
 70 |             temp = np.asarray([
 71 |                 xls_data['data'][i][9]-1,
 72 |                  xls_data['data'][i][10]-1,
 73 |                  xls_data['data'][i][11]-1,
 74 |                  xls_data['data'][i][12]-1,
 75 |                  xls_data['data'][i][13]-1,
 76 |                  xls_data['data'][i][14]-1,
 77 |                  xls_data['data'][i][8] +32])
 78 |             ssq_data.append(temp)
 79 |     else:
 80 |         for i in range(2, len(xls_data['data'])):
 81 |             # print(type(xls_data['data'][i][2]))
 82 |             # aaaa=xls_data['data'][i][2]
 83 |             if xls_data['data'][i] == []:
 84 |                 break
 85 |             temp = np.asarray([
 86 |                 xls_data['data'][i][2]-1,
 87 |                 xls_data['data'][i][3]-1,
 88 |                 xls_data['data'][i][4]-1,
 89 |                 xls_data['data'][i][5]-1,
 90 |                 xls_data['data'][i][6]-1,
 91 |                 xls_data['data'][i][7]-1,
 92 |                 xls_data['data'][i][8] +32])
 93 |             ssq_data.append(temp)
 94 |     return ssq_data
 95 | def get_exl_data_by_period(exl_path='./ssq.xls',random_order=False,use_resnet=False,times=10):
 96 |     ssq_data = []
 97 |     xls_data = get_data(exl_path)
 98 |     # print(type(xls_data))
 99 |     if use_resnet:
100 |         if random_order:
101 |             for i in range(2, len(xls_data['data'])-times):
102 |                 # print(type(xls_data['data'][i][2]))
103 |                 # aaaa=xls_data['data'][i][2]
104 |                 if xls_data['data'][i] == [] or xls_data['data'][i+times] == []:
105 |                     break
106 |                 temp=[]
107 |                 for j in range(times):
108 |                     temp.append([
109 |                     [xls_data['data'][i+j][9] - 1],
110 |                     [xls_data['data'][i+j][10] - 1],
111 |                     [xls_data['data'][i+j][11] - 1],
112 |                     [xls_data['data'][i+j][12] - 1],
113 |                     [xls_data['data'][i+j][13] - 1],
114 |                     [xls_data['data'][i+j][14] - 1],
115 |                     [xls_data['data'][i+j][8] + 32],
116 |                     [xls_data['data'][i+j][15] / 10000000000],
117 |                     [xls_data['data'][i+j][16] / 10000000000],
118 |                     [xls_data['data'][i+j][17] / 10000000],
119 |                     [xls_data['data'][i+j][19] / 10000000],
120 |                     [xls_data['data'][i+j][21] / 10000000],
121 |                     [xls_data['data'][i+j][23] / 10000000],
122 |                     [xls_data['data'][i+j][25] / 10000000],
123 |                     [xls_data['data'][i+j][27] / 10000000]
124 |                 ])
125 | 
126 |                 # print(temp.shape)
127 |                 #
128 |                 ssq_data.append(np.asarray(temp))
129 |         else:
130 |             for i in range(2, len(xls_data['data'])-times):
131 |                 # print(type(xls_data['data'][i][2]))
132 |                 # aaaa=xls_data['data'][i][2]
133 |                 if xls_data['data'][i] == [] or xls_data['data'][i+times] == []:
134 |                     break
135 |                 temp=[]
136 | 
137 |                 for j in range(times):
138 | 
139 |                     temp.append([
140 |                     [xls_data['data'][i+j][2] - 1],
141 |                     [xls_data['data'][i+j][3] - 1],
142 |                     [xls_data['data'][i+j][3] - 1],
143 |                     [xls_data['data'][i+j][5] - 1],
144 |                     [xls_data['data'][i+j][6] - 1],
145 |                     [xls_data['data'][i+j][7] - 1],
146 |                     [xls_data['data'][i+j][8] + 32],
147 |                     [xls_data['data'][i+j][15] / 10000000000],
148 |                     [xls_data['data'][i+j][16] / 10000000000],
149 |                     [xls_data['data'][i+j][17] / 10000000],
150 |                     [xls_data['data'][i+j][19] / 10000000],
151 |                     [xls_data['data'][i+j][21] / 10000000],
152 |                     [xls_data['data'][i+j][23] / 10000000],
153 |                     [xls_data['data'][i+j][25] / 10000000],
154 |                     [xls_data['data'][i+j][27] / 10000000]
155 |                 ])
156 |                 ssq_data.append(np.asarray(temp))
157 |         return ssq_data
158 | 
159 |     if random_order:
160 |         for i in range(2, len(xls_data['data'])):
161 |             # print(type(xls_data['data'][i][2]))
162 |             # aaaa=xls_data['data'][i][2]
163 |             if xls_data['data'][i] == []:
164 |                 break
165 |             temp = np.asarray([
166 |                 xls_data['data'][i][9] - 1,
167 |                 xls_data['data'][i][10] - 1,
168 |                 xls_data['data'][i][11] - 1,
169 |                 xls_data['data'][i][12] - 1,
170 |                 xls_data['data'][i][13] - 1,
171 |                 xls_data['data'][i][14] - 1,
172 |                 xls_data['data'][i][8] + 32])
173 |             ssq_data.append(temp)
174 |     else:
175 |         for i in range(2, len(xls_data['data'])):
176 |             # print(type(xls_data['data'][i][2]))
177 |             # aaaa=xls_data['data'][i][2]
178 |             if xls_data['data'][i] == []:
179 |                 break
180 |             temp = np.asarray([
181 |                 xls_data['data'][i][2] - 1,
182 |                 xls_data['data'][i][3] - 1,
183 |                 xls_data['data'][i][4] - 1,
184 |                 xls_data['data'][i][5] - 1,
185 |                 xls_data['data'][i][6] - 1,
186 |                 xls_data['data'][i][7] - 1,
187 |                 xls_data['data'][i][8] + 32])
188 |             ssq_data.append(temp)
189 |     return ssq_data
190 | 
191 | def get_red(exl_path='./ssq.xls', random_order=False):
192 |     ssq_data=[]
193 |     xls_data = get_data(exl_path)
194 |     # print(type(xls_data))
195 |     if random_order:
196 |         for i in range(2, len(xls_data['data'])):
197 |             # print(type(xls_data['data'][i][2]))
198 |             # aaaa=xls_data['data'][i][2]
199 |             temp = np.asarray([
200 |                 xls_data['data'][i][9]-1,
201 |                  xls_data['data'][i][10]-1,
202 |                  xls_data['data'][i][11]-1,
203 |                  xls_data['data'][i][12]-1,
204 |                  xls_data['data'][i][13]-1,
205 |                  xls_data['data'][i][14]-1])
206 |             ssq_data.append(temp)
207 |     else:
208 |         for i in range(2, len(xls_data['data'])):
209 |             # print(type(xls_data['data'][i][2]))
210 |             # aaaa=xls_data['data'][i][2]
211 |             temp = np.asarray([
212 |                 xls_data['data'][i][2] - 1,
213 |                 xls_data['data'][i][3] - 1,
214 |                 xls_data['data'][i][4] - 1,
215 |                 xls_data['data'][i][5] - 1,
216 |                 xls_data['data'][i][6] - 1,
217 |                 xls_data['data'][i][7] - 1])
218 |             ssq_data.append(temp)
219 |     return ssq_data
220 | 
221 | def get_blue(exl_path='./ssq.xls',use_resnet=False):
222 |     ssq_data=[]
223 |     xls_data = get_data(exl_path)
224 |     if use_resnet:
225 |         for i in range(2, len(xls_data['data'])):
226 |             # print(type(xls_data['data'][i][2]))
227 |             # aaaa=xls_data['data'][i][2]
228 |             temp = np.asarray([[[
229 |                 xls_data['data'][i][8] - 1]]])
230 |             ssq_data.append(temp)
231 |     else:
232 |         for i in range(2, len(xls_data['data'])):
233 |             # print(type(xls_data['data'][i][2]))
234 |             # aaaa=xls_data['data'][i][2]
235 |             temp = np.asarray([
236 |                 xls_data['data'][i][8]-1])
237 |             ssq_data.append(temp)
238 |     return ssq_data
239 | 
240 | def get_red161(exl_path='./ssq.xls',use_resnet=True):
241 |     ssq_data=[]
242 |     xls_data = get_data(exl_path)
243 |     # print(type(xls_data))
244 |     if use_resnet:
245 |         for i in range(2, len(xls_data['data'])):
246 |             # print(type(xls_data['data'][i][2]))
247 |             # aaaa=xls_data['data'][i][2]
248 |             temp = np.asarray([[
249 |                 [xls_data['data'][i][9]],
250 |                  [xls_data['data'][i][10]],
251 |                  [xls_data['data'][i][11]],
252 |                  [xls_data['data'][i][12]],
253 |                  [xls_data['data'][i][13]],
254 |                  [xls_data['data'][i][14]]]])
255 |             # print(temp.shape)
256 |             #
257 |             ssq_data.append(temp)
258 |         return ssq_data
259 | 
260 | def get_exl_data(exl_path='./ssq.xls',random_order=False,use_resnet=False):
261 |     ssq_data=[]
262 |     xls_data = get_data(exl_path)
263 |     # print(type(xls_data))
264 |     if use_resnet:
265 |         if random_order:
266 |             for i in range(2, len(xls_data['data'])):
267 |                 # print(type(xls_data['data'][i][2]))
268 |                 # aaaa=xls_data['data'][i][2]
269 |                 if xls_data['data'][i] == []:
270 |                     break
271 |                 temp = np.asarray([[
272 |                     [xls_data['data'][i][9]-1],
273 |                      [xls_data['data'][i][10]-1],
274 |                      [xls_data['data'][i][11]-1],
275 |                      [xls_data['data'][i][12]-1],
276 |                      [xls_data['data'][i][13]-1],
277 |                      [xls_data['data'][i][14]-1],
278 |                      [xls_data['data'][i][8]+32]]])
279 |                 # print(temp.shape)
280 |                 #
281 |                 ssq_data.append(temp)
282 |         else:
283 |             for i in range(2, len(xls_data['data'])):
284 |                 # print(type(xls_data['data'][i][2]))
285 |                 # aaaa=xls_data['data'][i][2]
286 |                 if xls_data['data'][i] == []:
287 |                     break
288 |                 temp = np.asarray([[
289 |                     [xls_data['data'][i][2] - 1],
290 |                     [xls_data['data'][i][3] - 1],
291 |                     [xls_data['data'][i][4] - 1],
292 |                     [xls_data['data'][i][5] - 1],
293 |                     [xls_data['data'][i][6] - 1],
294 |                     [xls_data['data'][i][7] - 1],
295 |                     [xls_data['data'][i][8] + 32]]])
296 |                 ssq_data.append(temp)
297 |         return ssq_data
298 | 
299 |     if random_order:
300 |         for i in range(2, len(xls_data['data'])):
301 |             # print(type(xls_data['data'][i][2]))
302 |             # aaaa=xls_data['data'][i][2]
303 |             if xls_data['data'][i] == []:
304 |                 break
305 |             temp = np.asarray([
306 |                 xls_data['data'][i][9]-1,
307 |                  xls_data['data'][i][10]-1,
308 |                  xls_data['data'][i][11]-1,
309 |                  xls_data['data'][i][12]-1,
310 |                  xls_data['data'][i][13]-1,
311 |                  xls_data['data'][i][14]-1,
312 |                  xls_data['data'][i][8] +32])
313 |             ssq_data.append(temp)
314 |     else:
315 |         for i in range(2, len(xls_data['data'])):
316 |             # print(type(xls_data['data'][i][2]))
317 |             # aaaa=xls_data['data'][i][2]
318 |             if xls_data['data'][i] == []:
319 |                 break
320 |             temp = np.asarray([
321 |                 xls_data['data'][i][2] - 1,
322 |                 xls_data['data'][i][3] - 1,
323 |                 xls_data['data'][i][4] - 1,
324 |                 xls_data['data'][i][5] - 1,
325 |                 xls_data['data'][i][6] - 1,
326 |                 xls_data['data'][i][7] - 1,
327 |                 xls_data['data'][i][8] +32])
328 |             ssq_data.append(temp)
329 |     return ssq_data
330 | 
331 | 
332 | 
333 | 
334 | def get_dlt_data(exl_path='./dlt.xls',random_order=False,use_resnet=False):
335 |     ssq_data=[]
336 |     xls_data = get_data(exl_path)
337 |     # print(type(xls_data))
338 |     if use_resnet:
339 |         if random_order:
340 |             for i in range(2, len(xls_data['data'])):
341 |                 # print(type(xls_data['data'][i][2]))
342 |                 # aaaa=xls_data['data'][i][2]
343 |                 if xls_data['data'][i] == []:
344 |                     break
345 |                 temp = np.asarray([[
346 |                     [xls_data['data'][i][9]-1],
347 |                      [xls_data['data'][i][10]-1],
348 |                      [xls_data['data'][i][11]-1],
349 |                      [xls_data['data'][i][12]-1],
350 |                      [xls_data['data'][i][13]-1],
351 |                      [xls_data['data'][i][14]-1],
352 |                      [xls_data['data'][i][8]+32]]])
353 |                 # print(temp.shape)
354 |                 #
355 |                 ssq_data.append(temp)
356 |         else:
357 |             for i in range(2, len(xls_data['data'])):
358 |                 # print(type(xls_data['data'][i][2]))
359 |                 # aaaa=xls_data['data'][i][2]
360 |                 if xls_data['data'][i] == []:
361 |                     break
362 |                 temp = np.asarray([[
363 |                     [xls_data['data'][i][2] - 1],
364 |                     [xls_data['data'][i][3] - 1],
365 |                     [xls_data['data'][i][4] - 1],
366 |                     [xls_data['data'][i][5] - 1],
367 |                     [xls_data['data'][i][6] - 1],
368 |                     [xls_data['data'][i][7] - 1],
369 |                     [xls_data['data'][i][8] + 32]]])
370 |                 ssq_data.append(temp)
371 |         return ssq_data
372 | 
373 |     if random_order:
374 |         for i in range(2, len(xls_data['data'])):
375 |             # print(type(xls_data['data'][i][2]))
376 |             # aaaa=xls_data['data'][i][2]
377 |             if xls_data['data'][i] == []:
378 |                 break
379 |             temp = np.asarray([
380 |                 xls_data['data'][i][9]-1,
381 |                  xls_data['data'][i][10]-1,
382 |                  xls_data['data'][i][11]-1,
383 |                  xls_data['data'][i][12]-1,
384 |                  xls_data['data'][i][13]-1,
385 |                  xls_data['data'][i][14]-1,
386 |                  xls_data['data'][i][8] +32])
387 |             ssq_data.append(temp)
388 |     else:
389 |         for i in range(2, len(xls_data['data'])):
390 |             # print(type(xls_data['data'][i][2]))
391 |             # aaaa=xls_data['data'][i][2]
392 |             if xls_data['data'][i] == []:
393 |                 break
394 |             temp = np.asarray([
395 |                 xls_data['data'][i][2] - 1,
396 |                 xls_data['data'][i][3] - 1,
397 |                 xls_data['data'][i][4] - 1,
398 |                 xls_data['data'][i][5] - 1,
399 |                 xls_data['data'][i][6] - 1,
400 |                 xls_data['data'][i][7] - 1,
401 |                 xls_data['data'][i][8] +32])
402 |             ssq_data.append(temp)
403 |     return ssq_data
404 | 
405 | 
406 | 
407 | def get_dlt_data(exl_path='./dlt2.xls',random_order=False,use_resnet=False):
408 |     ssq_data=[]
409 |     xls_data = get_data(exl_path)
410 |     # print(type(xls_data))
411 |     if use_resnet:
412 |         if random_order:
413 |             for i in range(2, len(xls_data['data'])):
414 |                 # print(type(xls_data['data'][i][2]))
415 |                 # aaaa=xls_data['data'][i][2]
416 |                 if xls_data['data'][i] == []:
417 |                     break
418 |                 temp = np.asarray([[
419 |                     [xls_data['data'][i][9]-1],
420 |                      [xls_data['data'][i][10]-1],
421 |                      [xls_data['data'][i][11]-1],
422 |                      [xls_data['data'][i][12]-1],
423 |                      [xls_data['data'][i][13]-1],
424 |                      [xls_data['data'][i][14]+34],
425 |                      [xls_data['data'][i][15]+34]]])
426 |                 # print(temp.shape)
427 |                 #
428 |                 ssq_data.append(temp)
429 |         else:
430 |             for i in range(2, len(xls_data['data'])):
431 |                 # print(type(xls_data['data'][i][2]))
432 |                 # aaaa=xls_data['data'][i][2]
433 |                 if xls_data['data'][i] == []:
434 |                     break
435 |                 temp = np.asarray([[
436 |                     [xls_data['data'][i][2] - 1],
437 |                     [xls_data['data'][i][3] - 1],
438 |                     [xls_data['data'][i][4] - 1],
439 |                     [xls_data['data'][i][5] - 1],
440 |                     [xls_data['data'][i][6] - 1],
441 |                     [xls_data['data'][i][7] + 34 ],
442 |                     [xls_data['data'][i][8] + 34]]])
443 |                 ssq_data.append(temp)
444 |         return ssq_data
445 | 
446 |     if random_order:
447 |         for i in range(2, len(xls_data['data'])):
448 |             # print(type(xls_data['data'][i][2]))
449 |             # aaaa=xls_data['data'][i][2]
450 |             if xls_data['data'][i] == []:
451 |                 break
452 |             temp = np.asarray([
453 |                 xls_data['data'][i][9]-1,
454 |                  xls_data['data'][i][10]-1,
455 |                  xls_data['data'][i][11]-1,
456 |                  xls_data['data'][i][12]-1,
457 |                  xls_data['data'][i][13]-1,
458 |                 xls_data['data'][i][14] + 34,
459 |                 xls_data['data'][i][15] + 34])
460 |             ssq_data.append(temp)
461 |     else:
462 |         for i in range(2, len(xls_data['data'])):
463 |             # print(type(xls_data['data'][i][2]))
464 |             # aaaa=xls_data['data'][i][2]
465 |             if xls_data['data'][i] == []:
466 |                 break
467 |             temp = np.asarray([
468 |                 xls_data['data'][i][2] - 1,
469 |                 xls_data['data'][i][3] - 1,
470 |                 xls_data['data'][i][4] - 1,
471 |                 xls_data['data'][i][5] - 1,
472 |                 xls_data['data'][i][6] - 1,
473 |                 xls_data['data'][i][7] + 34,
474 |                 xls_data['data'][i][8] + 34])
475 |             ssq_data.append(temp)
476 |     return ssq_data
477 | 


--------------------------------------------------------------------------------
/ssq_test.py:
--------------------------------------------------------------------------------
 1 | # -*- coding: utf-8 -*-
 2 | # file: main.py
 3 | # author: JinTian
 4 | # time: 11/03/2017 9:53 AM
 5 | # Copyright 2017 JinTian. All Rights Reserved.
 6 | #
 7 | # Licensed under the Apache License, Version 2.0 (the "License");
 8 | # you may not use this file except in compliance with the License.
 9 | # You may obtain a copy of the License at
10 | #
11 | #     http://www.apache.org/licenses/LICENSE-2.0
12 | #
13 | # Unless required by applicable law or agreed to in writing, software
14 | # distributed under the License is distributed on an "AS IS" BASIS,
15 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 | # See the License for the specific language governing permissions and
17 | # limitations under the License.
18 | # ------------------------------------------------------------------------
19 | import tensorflow as tf
20 | from poems.model import rnn_model
21 | from poems.poems import process_poems
22 | import numpy as np
23 | from ssq_data import *
24 | start_token = 'B'
25 | end_token = 'E'
26 | model_dir = './model/'
27 | corpus_file = './data/poems.txt'
28 | 
29 | 
30 | def to_word(predict, vocabs):
31 |     t = np.cumsum(predict)
32 |     s = np.sum(predict)
33 |     sample = int(np.searchsorted(t, np.random.rand(1) * s))
34 |     if sample > len(vocabs):
35 |         sample = len(vocabs) - 1
36 |     return vocabs[sample]
37 | 
38 | 
39 | def gen_poem():
40 |     batch_size = 1
41 |     print('## loading model from %s' % model_dir)
42 | 
43 |     input_data = tf.placeholder(tf.int32, [batch_size, None])
44 | 
45 |     end_points = rnn_model(model='lstm', input_data=input_data, output_data=None, vocab_size=33+16,
46 |                            rnn_size=128, output_num=7,input_num=7,num_layers=7, batch_size=1, learning_rate=0.01)
47 | 
48 |     saver = tf.train.Saver(tf.global_variables())
49 |     init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
50 |     with tf.Session() as sess:
51 |         sess.run(init_op)
52 | 
53 |         checkpoint = tf.train.latest_checkpoint(model_dir)
54 |         saver.restore(sess, checkpoint)
55 |         ssqdata = get_exl_data()
56 |         # x = np.array([list(map(word_int_map.get, start_token))])
57 |         x=[ssqdata[len(ssqdata)-1]]
58 |         print("input: %s"%(x+np.asarray([1,1,1,1,1,1,-32])))
59 |         [predict, last_state] = sess.run([end_points['prediction'], end_points['last_state']],
60 |                                          feed_dict={input_data: x})
61 |         poem_=np.argmax(np.array(predict),axis=1)
62 |         results=poem_+np.asarray([1,1,1,1,1,1,-32])
63 |         print("output:%s"%results)
64 |         return poem_
65 | 
66 | 
67 | 
68 | if __name__ == '__main__':
69 |     # begin_char = input('## please input the first character:')
70 |     poem = gen_poem()
71 |     # pretty_print_poem(poem_=poem)


--------------------------------------------------------------------------------