├── code ├── README ├── users_dataset.py ├── fully_connected.py ├── encoder_rnn.py ├── decoder_rnn.py ├── words_clustering.py ├── confusion_matrix_analysis.py ├── run_encoder.py ├── train_denoising_autoencoder.py ├── data_loading_preprocessing.py ├── plot_autoencoder.py ├── Bhattacharyya_distance.py ├── users_identification.py └── users_disambiguation.py ├── plots └── README ├── outputs └── README ├── input_files └── README ├── processed_files └── README ├── model_parameters └── README ├── README.md └── LICENSE /code/README: -------------------------------------------------------------------------------- 1 | Folder with the source code files. 2 | -------------------------------------------------------------------------------- /plots/README: -------------------------------------------------------------------------------- 1 | Folder in which the plots will be saved. 2 | -------------------------------------------------------------------------------- /outputs/README: -------------------------------------------------------------------------------- 1 | Folder in which the output files will be saved. 2 | -------------------------------------------------------------------------------- /input_files/README: -------------------------------------------------------------------------------- 1 | Folder in which the input files have to be saved. 2 | -------------------------------------------------------------------------------- /processed_files/README: -------------------------------------------------------------------------------- 1 | Folder in which the processed files will be saved. 2 | -------------------------------------------------------------------------------- /model_parameters/README: -------------------------------------------------------------------------------- 1 | Folder in which the the framework parameters will be saved. 2 | -------------------------------------------------------------------------------- /code/users_dataset.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | users_dataset: defines the dataset structure for the autoencoder 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | from torch.utils.data import Dataset 23 | 24 | 25 | class UsersDatasetAutoencoder(Dataset): 26 | 27 | def __init__(self, input_list): 28 | self.sentences = input_list 29 | 30 | def __len__(self): 31 | return len(self.sentences) 32 | 33 | def __getitem__(self, idx): 34 | sample = self.sentences[idx] 35 | return sample 36 | -------------------------------------------------------------------------------- /code/fully_connected.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | fully_connected: defines a fully connected layer with tanh activation function 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import torch 23 | import torch.nn as nn 24 | 25 | 26 | class FullyConnected(nn.Module): 27 | def __init__(self, hidden_size): 28 | super(FullyConnected, self).__init__() 29 | self.fc = nn.Linear(hidden_size, hidden_size) 30 | self.tan = nn.Tanh() 31 | 32 | def forward(self, x): 33 | x = torch.transpose(x, 0, 1) 34 | connect = self.fc(x) 35 | connect = self.tan(connect) 36 | connect = torch.transpose(connect, 0, 1) 37 | return connect 38 | -------------------------------------------------------------------------------- /code/encoder_rnn.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | encoder: defines the architecture of the GRU-based RNN encoder 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import torch.nn as nn 23 | 24 | 25 | class RNNEncoder(nn.Module): 26 | def __init__(self, hidden_size, num_layers): 27 | super(RNNEncoder, self).__init__() 28 | self.hidden_size = hidden_size 29 | self.num_layers = num_layers 30 | self.encoder = nn.GRU(1, hidden_size, num_layers) 31 | self.drop = nn.Dropout(p=0.2) 32 | 33 | def encode(self, x): 34 | _, hidden = self.encoder(x) 35 | hidden = self.drop(hidden) 36 | return hidden 37 | 38 | def forward(self, x): 39 | hidden = self.encode(x) 40 | return hidden 41 | -------------------------------------------------------------------------------- /code/decoder_rnn.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | decoder_rnn: defines the architecture of the GRU-based RNN decoder 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import torch 23 | import torch.nn as nn 24 | from torch.nn.utils.rnn import pad_packed_sequence 25 | 26 | 27 | class RNNDecoder(nn.Module): 28 | def __init__(self, hidden_size, num_layers): 29 | super(RNNDecoder, self).__init__() 30 | self.hidden_size = hidden_size 31 | self.num_layers = num_layers 32 | self.decoder = nn.GRU(1, hidden_size, num_layers) 33 | self.lin = nn.Linear(hidden_size, 1) 34 | 35 | def decode(self, x, hidden): 36 | decoder_output, hidden = self.decoder(x, hidden) 37 | decoder_output = pad_packed_sequence(decoder_output, batch_first=True)[0] 38 | decoder_output = torch.transpose(decoder_output, 0, 1) 39 | decoder_output = self.lin(decoder_output) 40 | decoder_output = torch.transpose(decoder_output, 0, 1) 41 | return decoder_output, hidden 42 | 43 | def forward(self, x, hidden): 44 | output, hidden = self.decode(x, hidden) 45 | return output, hidden 46 | -------------------------------------------------------------------------------- /code/words_clustering.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | sentences_clustering: clusters the codes into classes using the GMM-based clustering algorithm 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import argparse 23 | import numpy as np 24 | import matplotlib.pyplot as plt 25 | from sklearn.mixture import BayesianGaussianMixture 26 | import pickle 27 | plt.switch_backend('agg') 28 | 29 | if __name__ == '__main__': 30 | parser = argparse.ArgumentParser(description=__doc__) 31 | parser.add_argument('num_clusters', help='Number of classes for clustering', type=int) 32 | args = parser.parse_args() 33 | 34 | num_traces = 40 35 | 36 | # ---------------------------------------------------------------------------------------------------------------- # 37 | # SENTENCES CLUSTERING # 38 | 39 | hidden_sentences = [] 40 | lengths = [] 41 | for idx in range(num_traces): 42 | with open('../processed_files/hiddens_train_user' + str(idx) + '.txt', 43 | "rb") as fp: # Unpickling 44 | sentences_train = pickle.load(fp) 45 | hidden_sentences.extend(sentences_train.cpu().data.numpy()) 46 | lengths.append(len(sentences_train)) 47 | 48 | hidden_sentences = np.asarray(hidden_sentences) 49 | lengths = np.asarray(lengths) 50 | 51 | sentences = hidden_sentences 52 | 53 | num_clusters = args.num_clusters 54 | gauss = BayesianGaussianMixture(n_components=num_clusters, covariance_type='diag', max_iter=200).fit( 55 | sentences) 56 | 57 | with open('../processed_files/gauss.txt', "wb") as fp: # Pickling 58 | pickle.dump(gauss, fp) 59 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Smartphone identification via passive traffic fingerprinting 2 | This repository contains the reference code for the paper [''Smartphone Identification via Passive Traffic Fingerprinting: a Sequence-to-Sequence Learning Approach'' DOI: 10.1109/MNET.001.1900101](https://ieeexplore.ieee.org/document/9003304). 3 | 4 | If you find the project useful and you use this code, please cite our paper: 5 | ``` 6 | @article{Meneghello2020Network, 7 | author={Francesca Meneghello and Michele Rossi and Nicola Bui}, 8 | title={Smartphone Identification via Passive Traffic Fingerprinting: a Sequence-to-Sequence Learning Approach}, 9 | journal={IEEE Network}, 10 | volume={34}, 11 | number={2}, 12 | pages={112--120}, 13 | year={2020} 14 | } 15 | ``` 16 | 17 | # How to use 18 | Clone the repository and enter the folder with the python code: 19 | ```bash 20 | cd 21 | git clone https://github.com/francescamen/smartphone_identification 22 | cd code 23 | ``` 24 | 25 | Download the input data from http://researchdata.cab.unipd.it/id/eprint/292, unzip and put them into the ```input_files``` folder. 26 | 27 | ## Train and test the framework 28 | To create the smartphone fingerprints and uses them to correctly associate unknown traffic traces to the user labels execute the following commands: 29 | ```bash 30 | python data_loading_preprocessing.py 31 | python train_denoising_autoencoder.py 32 | python run_encoder.py 33 | python words_clustering.py 34 | python users_identification.py 35 | ``` 36 | ## Visualize the results 37 | In the ```code``` folder you can find other utilities functions. 38 | To visualize the performance of the autoencoder run the following command: 39 | ```bash 40 | python plot_autoencoder.py 41 | ``` 42 | The confusion matrix can be computed and plotted throught the command: 43 | ```bash 44 | python confusion_matrix_analysis.py 45 | ``` 46 | The users similarity assessment is performed by executing the following commands: 47 | ```bash 48 | python users_disambiguation.py 49 | python Bhattacharyya_distance.py 50 | ``` 51 | 52 | ## Parameters 53 | The results on the article are obtained with the following parameters: ```=32``` ```=2``` ```=100``` ```=50``` ```=100```. 54 | 55 | # Authors 56 | Francesca Meneghello, Michele Rossi, Nicola Bui 57 | 58 | # Contact 59 | meneghello@dei.unipd.it 60 | github.com/francescamen 61 | -------------------------------------------------------------------------------- /code/confusion_matrix_analysis.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | confusion_matrix_analysis: computes the confusion matrices 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import numpy as np 23 | import matplotlib.pyplot as plt 24 | 25 | if __name__ == '__main__': 26 | 27 | confusion_matrix = np.load('../outputs/confusion_matrix_test.npy') 28 | 29 | number_users = confusion_matrix.shape[0] 30 | 31 | precision = np.zeros((number_users, 1)) 32 | recall = np.zeros((number_users, 1)) 33 | f_score = np.zeros((number_users, 1)) 34 | true_positive = np.diag(confusion_matrix) 35 | false_positive = np.sum(confusion_matrix, axis=0) - true_positive 36 | false_negative = np.sum(confusion_matrix, axis=1) - true_positive 37 | for user in range(1, number_users + 1): 38 | if true_positive[user - 1] > 0: 39 | precision[user - 1] = true_positive[user - 1] / (true_positive[user - 1] + false_positive[user - 1]) 40 | recall[user - 1] = true_positive[user - 1] / (true_positive[user - 1] + false_negative[user - 1]) 41 | f_score[user - 1] = 2 * precision[user - 1] * recall[user - 1] / (precision[user - 1] + recall[user - 1]) 42 | else: 43 | precision[user - 1] = 0 44 | recall[user - 1] = 0 45 | f_score[user - 1] = 0 46 | 47 | f_score_mean = np.mean(f_score) 48 | 49 | confusion_matrix_normaliz_row = confusion_matrix / np.sum(confusion_matrix, axis=1).reshape(-1, 1) 50 | confusion_matrix_normaliz_column = \ 51 | confusion_matrix_normaliz_row / np.sum(confusion_matrix_normaliz_row, axis=0).reshape(1, -1) 52 | max_columns = np.amax(confusion_matrix_normaliz_column, axis=0) 53 | sum_max_columns = np.sum(max_columns) 54 | 55 | correct_windows = np.sum(np.diag(confusion_matrix)) 56 | number_windows = np.sum(np.sum(confusion_matrix, 1)) 57 | perc_correct_window = correct_windows/number_windows * 100 58 | print('perc_correct_windows: ' + str(perc_correct_window)) 59 | 60 | confusion_matrix_majority = np.zeros(confusion_matrix_normaliz_row.shape) 61 | correct = 0 62 | for r in range(0, confusion_matrix_normaliz_row.shape[0]): 63 | index_max = np.argmax(confusion_matrix_normaliz_row[r, :]) 64 | val_max = np.max(confusion_matrix_normaliz_row[r, :]) 65 | indices_maxs = [i for i, j in enumerate(confusion_matrix_normaliz_row[r, :]) if j == val_max] 66 | confusion_matrix_majority[r, indices_maxs] = 1/len(indices_maxs) 67 | if index_max == r and len(indices_maxs) == 1: 68 | correct = correct + 1 69 | perc_correct_user = correct/number_users * 100 70 | print('perc_correct_users: ' + str(perc_correct_user)) 71 | 72 | fig, axes = plt.subplots(nrows=1, ncols=2) 73 | fig.set_size_inches(12, 5) 74 | ax = axes.flat 75 | 76 | im1 = ax[0].pcolor(np.linspace(0.5, number_users + 0.5, number_users + 1), 77 | np.linspace(0.5, number_users + 0.5, number_users + 1), 78 | np.transpose(confusion_matrix_normaliz_row), 79 | cmap='Blues', edgecolors='black', vmin=0, vmax=1) 80 | ax[0].set_title(r"$\bf{" + "Normalized" + "}$" + " " r"$\bf{" + "confusion" + "}$" + " " r"$\bf{" + "matrix" + "}$", 81 | FontSize=14) 82 | ax[0].set_xlabel('Actual user', FontSize=14) 83 | ax[0].set_xticks(np.linspace(1, number_users, number_users), minor=True) 84 | ax[0].set_yticks(np.linspace(1, number_users, number_users), minor=True) 85 | ax[0].set_ylabel('Predicted user', FontSize=14) 86 | ax[0].tick_params(axis="x", labelsize=11) 87 | ax[0].tick_params(axis="y", labelsize=11) 88 | 89 | im2 = ax[1].pcolor(np.linspace(0.5, number_users + 0.5, number_users + 1), 90 | np.linspace(0.5, number_users + 0.5, number_users + 1), np.transpose(confusion_matrix_majority), 91 | cmap='Blues', edgecolors='black', vmin=0, vmax=1) 92 | 93 | ax[1].set_title(r"$\bf{" + "Majority" + "}$" + " " r"$\bf{" + "score" + "}$", FontSize=14) 94 | ax[1].set_xlabel('Actual user', FontSize=14) 95 | ax[1].set_xticks(np.linspace(1, number_users, number_users), minor=True) 96 | ax[1].set_yticks(np.linspace(1, number_users, number_users), minor=True) 97 | ax[1].set_ylabel('Predicted user', FontSize=14) 98 | ax[1].tick_params(axis="x", labelsize=11) 99 | ax[1].tick_params(axis="y", labelsize=11) 100 | 101 | fig.subplots_adjust(right=0.88) 102 | cbar_ax = fig.add_axes([0.90, 0.15, 0.02, 0.7]) 103 | cbar = fig.colorbar(im1, cax=cbar_ax) 104 | cbar.ax.set_ylabel('Accuracy', FontSize=14) 105 | cbar.ax.tick_params(axis="y", labelsize=11) 106 | 107 | plt.savefig('../plots/cm_total.eps') 108 | 109 | plt.show() 110 | -------------------------------------------------------------------------------- /code/run_encoder.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | run_encoder: outputs the code for each input word using the GRU-based RNN encoder 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import argparse 23 | import numpy as np 24 | import matplotlib.pyplot as plt 25 | import torch 26 | from encoder_rnn import RNNEncoder 27 | from fully_connected import FullyConnected 28 | from decoder_rnn import RNNDecoder 29 | import pickle 30 | plt.switch_backend('agg') 31 | device = torch.device("cuda") 32 | 33 | 34 | def tensor_from_sentence(sentence): 35 | sent_list = list(sentence) 36 | return torch.tensor(sent_list).view(-1, 1).float() 37 | 38 | 39 | if __name__ == '__main__': 40 | parser = argparse.ArgumentParser(description=__doc__) 41 | parser.add_argument('hidden_neurons', help='Number of hidden neurons', type=int) 42 | parser.add_argument('layers', help='Number of layers', type=int) 43 | args = parser.parse_args() 44 | 45 | num_traces = 40 46 | 47 | with open('../processed_files/mInfo_train.txt', "rb") as fp: # Unpickling 48 | mInfo_train = pickle.load(fp) 49 | with open('../processed_files/mInfo_test.txt', "rb") as fp: # Unpickling 50 | mInfo_test = pickle.load(fp) 51 | 52 | with open('../processed_files/mTime_train.txt', "rb") as fp: # Unpickling 53 | mTime_train = pickle.load(fp) 54 | with open('../processed_files/mTime_test.txt', "rb") as fp: # Unpickling 55 | mTime_test = pickle.load(fp) 56 | 57 | with open('../processed_files/sentences_train.txt', "rb") as fp: # Unpickling 58 | sentences_train = pickle.load(fp) 59 | with open('../processed_files/sentences_test.txt', "rb") as fp: # Unpickling 60 | sentences_test = pickle.load(fp) 61 | 62 | # ---------------------------------------------------------------------------------------------------------------- # 63 | # CODE THE SENTENCES # 64 | 65 | encoder_model = RNNEncoder(args.hidden_neurons, args.layers).to(device) 66 | fully_connect = FullyConnected(args.hidden_neurons).to(device) 67 | 68 | encoder_model.load_state_dict(torch.load('../model_parameters/encoder_model_' + str(num_traces) + '.pt')) 69 | fully_connect.load_state_dict(torch.load('../model_parameters/fully_connected_' + str(num_traces) + '.pt')) 70 | 71 | encoder_model.eval() 72 | fully_connect.eval() 73 | 74 | mTime_train = np.asarray(mTime_train) 75 | mTime_test = np.asarray(mTime_test) 76 | mInfo_train = np.asarray(mInfo_train) 77 | mInfo_test = np.asarray(mInfo_test) 78 | 79 | for trace in range(num_traces): 80 | indices = np.argwhere(mInfo_train[:, 7] == trace + 1)[:, 0] 81 | sentences_single_user_train = sentences_train[indices[0]:indices[-1]] 82 | 83 | hiddens = torch.zeros((len(sentences_single_user_train), args.hidden_neurons*args.layers), device=device) 84 | for i_w in range(len(sentences_single_user_train)): 85 | x_input = tensor_from_sentence(sentences_single_user_train[i_w]) 86 | x_input = torch.reshape(x_input, (x_input.shape[0], 1, 1)).to(device) 87 | representation = fully_connect(encoder_model(x_input)) 88 | representation = torch.transpose(representation, 0, 1).contiguous().to(device) 89 | representation = representation.view(1, -1) 90 | hiddens[i_w, :] = representation 91 | 92 | with open('../processed_files/hiddens_train_user' + str(trace) + '.txt', "wb") as fp: # Pickling 93 | pickle.dump(hiddens, fp) 94 | with open('../processed_files/times_train_user' + str(trace) + '.txt', "wb") as fp: # Pickling 95 | pickle.dump(mTime_train[indices[0]:indices[-1], :], fp) 96 | 97 | indices = np.argwhere(mInfo_test[:, 7] == trace + 1)[:, 0] 98 | sentences_single_user_test = sentences_test[indices[0]:indices[-1]] 99 | 100 | hiddens = torch.zeros((len(sentences_single_user_test), args.hidden_neurons*args.layers)) 101 | for i_w in range(len(sentences_single_user_test)): 102 | x_input = tensor_from_sentence(sentences_single_user_test[i_w]) 103 | x_input = torch.reshape(x_input, (x_input.shape[0], 1, 1)).to(device) 104 | representation = fully_connect(encoder_model(x_input)) 105 | representation = torch.transpose(representation, 0, 1).contiguous().to(device) 106 | representation = representation.view(1, -1) 107 | hiddens[i_w, :] = representation 108 | 109 | with open('../processed_files/hiddens_test_user' + str(trace) + '.txt', "wb") as fp: # Pickling 110 | pickle.dump(hiddens, fp) 111 | with open('../processed_files/times_test_user' + str(trace) + '.txt', "wb") as fp: # Pickling 112 | pickle.dump(mTime_test[indices[0]:indices[-1], :], fp) 113 | -------------------------------------------------------------------------------- /code/train_denoising_autoencoder.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | train_denoising_autoencoder: trains the autoencoder and save the coding network parameters 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import argparse 23 | import numpy as np 24 | import matplotlib.pyplot as plt 25 | import torch 26 | import torch.nn as nn 27 | import torch.optim as optim 28 | from torch.utils.data import DataLoader 29 | from torch.nn.utils.rnn import pack_sequence 30 | from torch.nn.utils.rnn import pad_packed_sequence 31 | from encoder_rnn import RNNEncoder 32 | from fully_connected import FullyConnected 33 | from decoder_rnn import RNNDecoder 34 | from users_dataset import UsersDatasetAutoencoder 35 | import pickle 36 | import gc 37 | plt.switch_backend('agg') 38 | device = torch.device("cuda") 39 | 40 | 41 | def tensor_from_sentence(sentence): 42 | return torch.tensor(list(sentence), device=device).view(-1, 1).float() 43 | 44 | 45 | def collate_fn_autoencoder(_list): 46 | _list.sort(key=len, reverse=True) 47 | return _list 48 | 49 | 50 | def collate_fn(_list): 51 | return _list 52 | 53 | 54 | def input_packing(_list): 55 | tensor_list = [tensor_from_sentence(_list[i]) for i in range(len(_list))] 56 | return pack_sequence(tensor_list) 57 | 58 | 59 | def target_packing(_list): 60 | tensor_list = [tensor_from_sentence(torch.cat((torch.zeros((1,), device=device), _list[i][:-1, 0]), 0)) for i in 61 | range(len(_list))] 62 | packed_list = pack_sequence(tensor_list) 63 | return packed_list 64 | 65 | 66 | if __name__ == '__main__': 67 | parser = argparse.ArgumentParser(description=__doc__) 68 | parser.add_argument('hidden_neurons', help='Number of hidden neurons', type=int) 69 | parser.add_argument('layers', help='Number of layers', type=int) 70 | parser.add_argument('epochs', help='Number of epochs', type=int) 71 | args = parser.parse_args() 72 | 73 | num_traces = 40 74 | 75 | with open('../processed_files/sentences_train.txt', "rb") as fp: # Unpickling 76 | sentences_train = pickle.load(fp) 77 | with open('../processed_files/sentences_test.txt', "rb") as fp: # Unpickling 78 | sentences_test = pickle.load(fp) 79 | 80 | # ---------------------------------------------------------------------------------------------------------------- # 81 | # AUTOENCODER # 82 | 83 | encoder_model = RNNEncoder(args.hidden_neurons, args.layers).to(device) 84 | fully_connect = FullyConnected(args.hidden_neurons).to(device) 85 | decoder_model = RNNDecoder(args.hidden_neurons, args.layers).to(device) 86 | 87 | print(encoder_model) 88 | print(fully_connect) 89 | print(decoder_model) 90 | 91 | criterion = nn.L1Loss(reduction='none') 92 | criterion_val = nn.L1Loss() 93 | 94 | parameters_dictionary = list(encoder_model.parameters()) + list(fully_connect.parameters()) + \ 95 | list(decoder_model.parameters()) 96 | 97 | optimizer = optim.Adam(parameters_dictionary, lr=0.001) 98 | 99 | tensor_sentences_train = [tensor_from_sentence(sentences_train[i]) for i in range(len(sentences_train))] 100 | users_dataset_train = UsersDatasetAutoencoder(tensor_sentences_train) 101 | 102 | tensor_sentences_test = [tensor_from_sentence(sentences_test[i]) for i in range(len(sentences_test))] 103 | users_dataset_test = UsersDatasetAutoencoder(tensor_sentences_test) 104 | 105 | batch_size = 64 106 | train_loader = DataLoader(users_dataset_train, batch_size=batch_size, shuffle=True, num_workers=0, 107 | collate_fn=collate_fn_autoencoder) 108 | 109 | test_loader = DataLoader(users_dataset_test, batch_size=batch_size, shuffle=True, num_workers=0, 110 | collate_fn=collate_fn_autoencoder) 111 | 112 | max_epochs = args.epochs 113 | print_loss_total = 0 114 | print_every = 500 115 | print_every_val = 50 116 | iteration = 0 117 | for epoch in range(max_epochs): 118 | print('----epoch ' + str(epoch) + ' training----') 119 | for i_batch, sample_batched in enumerate(train_loader): 120 | iteration += 1 121 | x_input = input_packing(sample_batched) 122 | x_target = target_packing(sample_batched).to(device) 123 | 124 | erasure_prob = 0.1 125 | sample_batched_noise = [] 126 | for ii in range(len(sample_batched)): 127 | random_mask_array = np.random.random(sample_batched[ii].shape[0]) 128 | random_mask_array = random_mask_array > erasure_prob 129 | random_mask_array = random_mask_array.astype(int) 130 | random_mask = torch.tensor(random_mask_array, device=device).float().view(-1, 1) 131 | sample_batched_noise.append(sample_batched[ii] * random_mask) 132 | x_input_noise = input_packing(sample_batched_noise).to(device) 133 | 134 | encoder_model.zero_grad() 135 | fully_connect.zero_grad() 136 | decoder_model.zero_grad() 137 | 138 | hidden = encoder_model(x_input_noise).to(device) 139 | 140 | conn = fully_connect(hidden).contiguous().to(device) 141 | 142 | out, dec_hidden = decoder_model(x_target, conn) 143 | 144 | output_unpacked = out 145 | 146 | input_unpacked = pad_packed_sequence(x_input, batch_first=True) 147 | mask = torch.clone(input_unpacked[0]) 148 | mask[mask > 0] = 1 149 | output_unpacked = output_unpacked * mask 150 | 151 | loss = criterion(output_unpacked, input_unpacked[0]) 152 | 153 | loss = loss 154 | loss_array = loss.data.cpu().numpy() 155 | loss_sum = torch.sum(loss, 1) 156 | lengths = input_unpacked[1].to(device) 157 | lengths_array = lengths.data.cpu().numpy() 158 | loss_mean = loss_sum[:, 0] / lengths.float() 159 | loss = torch.sum(loss_mean) 160 | 161 | loss.backward() 162 | optimizer.step() 163 | 164 | print_loss_total += loss 165 | if iteration % print_every == 0: 166 | print_loss_avg = print_loss_total / print_every 167 | print_loss_total = 0 168 | print(print_loss_avg) 169 | 170 | del x_input 171 | del x_target 172 | del out 173 | del output_unpacked 174 | del lengths 175 | del dec_hidden 176 | gc.collect() 177 | 178 | print('----epoch ' + str(epoch) + ' test----') 179 | iteration = 0 180 | print_loss_total_test = 0 181 | for i_batch, sample_batched in enumerate(test_loader): 182 | iteration += 1 183 | x_input = input_packing(sample_batched).to(device) 184 | 185 | hidden = encoder_model(x_input).to(device) 186 | conn = fully_connect(hidden).contiguous().to(device) 187 | 188 | out_list = [] 189 | loss_mean_test = torch.zeros(batch_size, device=device) 190 | for bat in range(len(sample_batched)): 191 | dec_out = [torch.zeros(1, 1, 1, device=device)] 192 | out_vector = torch.zeros(sample_batched[bat].shape[0], 1, device=device) 193 | dec_hidden = conn[:, bat:bat + 1, :].contiguous() 194 | for word in range(out_vector.shape[0]): 195 | dec_out = input_packing(dec_out).to(device) 196 | out, dec_hidden = decoder_model(dec_out, dec_hidden) 197 | dec_hidden = dec_hidden.contiguous() 198 | dec_out = out.contiguous() 199 | out_vector[word] = dec_out 200 | out_list.append(out_vector) 201 | loss_mean_test[bat] = criterion_val(out_vector, sample_batched[bat]) 202 | loss_mean_test_array = loss_mean_test.cpu().detach().numpy() 203 | loss_test = torch.sum(loss_mean_test) 204 | 205 | print_loss_total_test += loss_test 206 | if iteration % print_every_val == 0: 207 | print_loss_avg_test = print_loss_total_test / print_every_val 208 | print_loss_total_test = 0 209 | print(print_loss_avg_test) 210 | 211 | del out_list 212 | gc.collect() 213 | 214 | torch.save(encoder_model.state_dict(), '../model_parameters/encoder_model_' + str(num_traces) + '.pt') 215 | torch.save(decoder_model.state_dict(), '../model_parameters/decoder_model_' + str(num_traces) + '.pt') 216 | torch.save(fully_connect.state_dict(), '../model_parameters/fully_connected_' + str(num_traces) + 217 | '.pt') 218 | -------------------------------------------------------------------------------- /code/data_loading_preprocessing.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | data_loading_preprocessing: loads the Matlab data and outputs the files to be used for further processing 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import numpy as np 23 | import matplotlib.pyplot as plt 24 | import scipy.io as sio 25 | import pickle 26 | plt.switch_backend('agg') 27 | 28 | 29 | if __name__ == '__main__': 30 | num_traces = 40 31 | 32 | # ---------------------------------------------------------------------------------------------------------------- # 33 | # FILE LOADING # 34 | 35 | # Load the Matlab files 36 | mPack_structure = sio.loadmat('../input_files/mPack.mat') 37 | mPack = mPack_structure['mPack'] 38 | 39 | mTime_structure = sio.loadmat('../input_files/mTime.mat') 40 | mTime = mTime_structure['mTime'] 41 | 42 | mInfo_structure = sio.loadmat('../input_files/mInfo.mat') 43 | mInfo = mInfo_structure['mInfo'] 44 | 45 | validInt_struct = sio.loadmat('../input_files/validInts.mat') 46 | validInts = validInt_struct['validInts'] 47 | 48 | # Order the file with respect to the absolute time, ordering inside each trace 49 | traceNum = 1 50 | mTimeSelect = [] 51 | mPackSelect = [] 52 | mInfoSelect = [] 53 | for trace in range(1, num_traces + 1): 54 | indices = np.argwhere(mInfo[:, -1] == trace) 55 | 56 | mTime_trace = mTime[indices[:, 0], :] 57 | mPack_trace = mPack[indices[:, 0], :] 58 | mInfo_trace = mInfo[indices[:, 0], :] 59 | 60 | mInfo_trace[:, -1] = traceNum # to label each selected trace with increasing indices 61 | traceNum = traceNum + 1 62 | 63 | # concatenate all the information in a single vector in order to order all the information in the same manner 64 | array_to_be_ordered = np.zeros((mTime_trace.shape[0], mTime_trace.shape[1] + mPack_trace.shape[1] + 65 | mInfo_trace.shape[1])) 66 | array_to_be_ordered[:, 0:mTime_trace.shape[1]] = mTime_trace 67 | array_to_be_ordered[:, mTime_trace.shape[1]:mTime_trace.shape[1] + mPack_trace.shape[1]] = mPack_trace 68 | array_to_be_ordered[:, mTime_trace.shape[1] + mPack_trace.shape[1]:mTime_trace.shape[1] + mPack_trace.shape[1] 69 | + mInfo_trace.shape[1]] = mInfo_trace 70 | 71 | # order the array 72 | array_ordered = array_to_be_ordered[array_to_be_ordered[:, 1].argsort()] 73 | 74 | # extract each singular information from the vector 75 | mTime_trace = array_ordered[:, 0:mTime_trace.shape[1]] 76 | mPack_trace = array_ordered[:, mTime_trace.shape[1]:mTime_trace.shape[1] + mPack_trace.shape[1]] 77 | mInfo_trace = array_ordered[:, mTime_trace.shape[1] + mPack_trace.shape[1]:mTime_trace.shape[1] 78 | + mPack_trace.shape[1] 79 | + mInfo_trace.shape[1]] 80 | 81 | # save the information in the matrices with all the traces 82 | mTimeSelect.append(mTime_trace) 83 | mPackSelect.append(mPack_trace) 84 | mInfoSelect.append(mInfo_trace) 85 | mTime = np.vstack(mTimeSelect) 86 | mPack = np.vstack(mPackSelect) 87 | mInfo = np.vstack(mInfoSelect) 88 | 89 | # Cancel the sentences with a length smaller than threshold, deleting also all the related information 90 | # to that aim, first concatenate all the information in a single vector 91 | array = np.zeros((mTime.shape[0], mTime.shape[1] + mPack.shape[1] + mInfo.shape[1])) 92 | array[:, 0:mTime.shape[1]] = mTime 93 | array[:, mTime.shape[1]:mTime.shape[1] + mPack.shape[1]] = mPack 94 | array[:, mTime.shape[1] + mPack.shape[1]:mTime.shape[1] + mPack.shape[1] + mInfo.shape[1]] = mInfo 95 | 96 | # create a new array in which to insert only the sentences with more than threshold samples 97 | new_array = np.zeros((mTime.shape[0], mTime.shape[1] + mPack.shape[1] + mInfo.shape[1])) 98 | index = 0 99 | threshold = 6 # the words smaller than that threshold will be discarded 100 | for i in range(array.shape[0]): 101 | if (array[i, mTime.shape[1] + 2 + threshold] != 0) or (np.sum(array[i, mTime.shape[1] + 2: mTime.shape[1] + 2 102 | + threshold]) > 500): 103 | new_array[index, :] = array[i, :] 104 | index = index + 1 105 | new_array = new_array[:index, :] 106 | 107 | # Select only the patterns inside the valid intervals extracted with Matlab 108 | new_array_2 = np.zeros((new_array.shape[0], mTime.shape[1] + mPack.shape[1] + mInfo.shape[1])) 109 | index = 0 110 | for i in range(new_array.shape[0]): 111 | trace_index = int(new_array[i, mPack.shape[1] + mTime.shape[1] + mInfo.shape[1] - 1]) 112 | start_valid_time = validInts[0][trace_index - 1][0, 0] + 5*60 113 | end_valid_time = validInts[0][trace_index - 1][0, 1] 114 | if end_valid_time > 37000: # cut all traces at 10 hours 115 | end_valid_time = start_valid_time + 36030 116 | if (new_array[i, 1] > start_valid_time) and (new_array[i, 1] < end_valid_time): 117 | new_array_2[index, :] = new_array[i, :] 118 | index = index + 1 119 | new_array_2 = new_array_2[:index, :] 120 | 121 | mTime_new = new_array_2[:, 0:mTime.shape[1]] 122 | mPack_new = new_array_2[:, mTime.shape[1]:mTime.shape[1] + mPack.shape[1]] 123 | mInfo_new = new_array_2[:, mTime.shape[1] + mPack.shape[1]:mTime.shape[1] + mPack.shape[1] + mInfo.shape[1]] 124 | 125 | trainFraction = 0.8 126 | sentences_train_pad = [] 127 | sentences_test_pad = [] 128 | mInfo_train = [] 129 | mInfo_test = [] 130 | mTime_train = [] 131 | mTime_test = [] 132 | dir_train = [] 133 | dir_test = [] 134 | 135 | for trace in range(1, num_traces + 1): 136 | indices = np.argwhere(mInfo_new[:, -1] == trace) 137 | 138 | mPack_trace = mPack_new[indices[:, 0], :] 139 | mInfo_trace = mInfo_new[indices[:, 0], :] 140 | mTime_trace = mTime_new[indices[:, 0], :] 141 | 142 | time_ass = mTime_trace[:, 1] - mTime_trace[0, 1] 143 | length_user_time = time_ass[-1] 144 | trLen_time = int(length_user_time*trainFraction) 145 | 146 | trLen = np.argwhere(time_ass < trLen_time)[-1, 0] 147 | 148 | sentences_train_pad.extend(mPack_trace[:trLen, 2:]/20000) 149 | sentences_test_pad.extend(mPack_trace[trLen:, 2:]/20000) 150 | 151 | mInfo_train.extend(mInfo_trace[:trLen, :]) 152 | mInfo_test.extend(mInfo_trace[trLen:, :]) 153 | 154 | mTime_train.extend(mTime_trace[:trLen, :]) 155 | mTime_test.extend(mTime_trace[trLen:, :]) 156 | 157 | dir_train.extend(mPack_trace[:trLen, 0]) 158 | dir_test.extend(mPack_trace[trLen:, 0]) 159 | 160 | sentences_train = [] 161 | sentences_test = [] 162 | for i in range(len(sentences_train_pad)): 163 | idx = np.argwhere(sentences_train_pad[i] == 0) 164 | if idx.size != 0: 165 | sentences_train.append(sentences_train_pad[i][:idx[0, 0]]) 166 | else: 167 | sentences_train.append(sentences_train_pad[i]) 168 | 169 | for i in range(len(sentences_test_pad)): 170 | idx = np.argwhere(sentences_test_pad[i] == 0) 171 | if idx.size != 0: 172 | sentences_test.append(sentences_test_pad[i][:idx[0, 0]]) 173 | else: 174 | sentences_test.append(sentences_test_pad[i]) 175 | 176 | del sentences_train_pad 177 | del sentences_test_pad 178 | del mInfo 179 | del mTime 180 | del mPack 181 | del mInfo_new 182 | del mPack_new 183 | del mTime_new 184 | del array 185 | del new_array_2 186 | del new_array 187 | del validInts 188 | del validInt_struct 189 | del mPack_structure 190 | del mInfo_structure 191 | del mTime_structure 192 | 193 | with open('../processed_files/mInfo_train.txt', "wb") as fp: # Pickling 194 | pickle.dump(mInfo_train, fp) 195 | with open('../processed_files/mInfo_test.txt', "wb") as fp: # Pickling 196 | pickle.dump(mInfo_test, fp) 197 | 198 | with open('../processed_files/mTime_train.txt', "wb") as fp: # Pickling 199 | pickle.dump(mTime_train, fp) 200 | with open('../processed_files/mTime_test.txt', "wb") as fp: # Pickling 201 | pickle.dump(mTime_test, fp) 202 | 203 | with open('../processed_files/sentences_train.txt', "wb") as fp: # Pickling 204 | pickle.dump(sentences_train, fp) 205 | with open('../processed_files/sentences_test.txt', "wb") as fp: # Pickling 206 | pickle.dump(sentences_test, fp) 207 | -------------------------------------------------------------------------------- /code/plot_autoencoder.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | plot_autoencoder: tests the autoencoder performance and plots the results 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import argparse 23 | import numpy as np 24 | import matplotlib 25 | import matplotlib.pyplot as plt 26 | import torch 27 | from torch.utils.data import DataLoader 28 | from torch.nn.utils.rnn import pack_sequence 29 | from torch.nn.utils.rnn import pad_sequence 30 | from encoder_rnn import RNNEncoder 31 | from fully_connected import FullyConnected 32 | from decoder_rnn import RNNDecoder 33 | from users_dataset import UsersDatasetAutoencoder 34 | import matplotlib.gridspec as gridspec 35 | import pickle 36 | matplotlib.use('TkAgg') 37 | 38 | 39 | def tensor_from_sentence(sentence): 40 | sent_list = list(sentence) 41 | return torch.tensor(sent_list).view(-1, 1).float() 42 | 43 | 44 | def collate_fn(_list): 45 | _list.sort(key=lambda x: x[0].shape[0], reverse=True) 46 | return _list 47 | 48 | 49 | def input_packing(_list): 50 | tensor_list = [tensor_from_sentence(_list[i]) for i in range(len(_list))] 51 | packed_list = pack_sequence(tensor_list) 52 | return packed_list 53 | 54 | 55 | if __name__ == '__main__': 56 | parser = argparse.ArgumentParser(description=__doc__) 57 | parser.add_argument('hidden_neurons', help='Number of hidden neurons', type=int) 58 | parser.add_argument('layers', help='Number of layers', type=int) 59 | args = parser.parse_args() 60 | 61 | num_traces = 40 62 | 63 | with open('../processed_files/mInfo_train.txt', "rb") as fp: # Unpickling 64 | mInfo_train = pickle.load(fp) 65 | with open('../processed_files/mInfo_test.txt', "rb") as fp: # Unpickling 66 | mInfo_test = pickle.load(fp) 67 | 68 | with open('../processed_files/mTime_train.txt', "rb") as fp: # Unpickling 69 | mTime_train = pickle.load(fp) 70 | with open('../processed_files/mTime_test.txt', "rb") as fp: # Unpickling 71 | mTime_test = pickle.load(fp) 72 | 73 | with open('../processed_files/sentences_train.txt', "rb") as fp: # Unpickling 74 | sentences_train = pickle.load(fp) 75 | with open('../processed_files/sentences_test.txt', "rb") as fp: # Unpickling 76 | sentences_test = pickle.load(fp) 77 | encoder_model = RNNEncoder(args.hidden_neurons, args.layers) 78 | fully_connect = FullyConnected(args.hidden_neurons) 79 | decoder_model = RNNDecoder(args.hidden_neurons, args.layers) 80 | 81 | encoder_model.load_state_dict(torch.load('../model_parameters/encoder_model_' + str(num_traces) + '.pt')) 82 | fully_connect.load_state_dict(torch.load('../model_parameters/fully_connected_' + str(num_traces) + '.pt')) 83 | decoder_model.load_state_dict(torch.load('../model_parameters/decoder_model_' + str(num_traces) + '.pt')) 84 | 85 | tensor_sentences_train = [[tensor_from_sentence(sentences_train[i]), tensor_from_sentence(mInfo_train[i][7:8] - 1), 86 | tensor_from_sentence(mTime_train[i])] for i in range(len(sentences_train))] 87 | users_dataset_train = UsersDatasetAutoencoder(tensor_sentences_train) 88 | 89 | tensor_sentences_test = [[tensor_from_sentence(sentences_test[i]), 90 | tensor_from_sentence(mInfo_test[i][7:8] - 1), 91 | tensor_from_sentence(mTime_test[i])] for i in range(len(sentences_test))] 92 | users_dataset_test = UsersDatasetAutoencoder(tensor_sentences_test) 93 | 94 | batch_size = 1 95 | train_loader = DataLoader(users_dataset_train, batch_size=batch_size, shuffle=False, num_workers=0, 96 | collate_fn=collate_fn) 97 | 98 | test_loader = DataLoader(users_dataset_test, batch_size=batch_size, shuffle=False, num_workers=0, 99 | collate_fn=collate_fn) 100 | 101 | fig = plt.figure(1) 102 | gridspec.GridSpec(2, 1) 103 | plt.subplot2grid((2, 1), (0, 0)) 104 | 105 | index_last = 0 106 | for i in range(50, 60): 107 | sample_batched = [users_dataset_test[i]] 108 | sample_batched_sent = [sample_batched[i][0] for i in range(len(sample_batched))] 109 | if sample_batched_sent[0].shape[0] > 0: 110 | sample_batched_labels = [sample_batched[i][1] for i in range(len(sample_batched))] 111 | sample_batched_time = [sample_batched[i][2][1] for i in range(len(sample_batched))] 112 | sample_batched_duration = [sample_batched[i][2][0]*10 for i in range(len(sample_batched))] 113 | 114 | sample_batched_sent = torch.stack(sample_batched_sent) 115 | sample_batched_sent = torch.transpose(sample_batched_sent, 0, 1) 116 | x_input = sample_batched_sent 117 | x_input = torch.reshape(x_input, (x_input.shape[0], 1, 1)) 118 | hidden_layer = fully_connect(encoder_model(x_input)) 119 | representation = torch.transpose(hidden_layer, 0, 1) 120 | representation = representation.view(1, -1) 121 | 122 | dec_out = [torch.zeros(1, 1, 1)] 123 | out_vector = torch.zeros(sample_batched_sent.shape[0], 1) 124 | dec_hidden = hidden_layer[:, :, :] 125 | for word in range(out_vector.shape[0]): 126 | dec_out = input_packing(dec_out) 127 | out, dec_hidden = decoder_model(dec_out, dec_hidden) 128 | dec_out = out 129 | out_vector[word] = dec_out 130 | 131 | out_vector_array = pad_sequence(out_vector).data[0].numpy() 132 | x_input_array = x_input.data.numpy()[:, 0, 0] 133 | 134 | x_ax = np.linspace(index_last, 135 | out_vector_array.shape[0] + index_last, 136 | out_vector_array.shape[0]) 137 | index_last = index_last + out_vector_array.shape[0] 138 | 139 | plt.plot(x_ax, out_vector_array*20000, c='firebrick', marker='x', linestyle='--', markersize=3, 140 | linewidth=0.8) 141 | plt.plot(x_ax, x_input_array*20000, c='midnightblue', marker='o', markersize=3, linewidth=0.8) 142 | plt.axvline(x=index_last + 1, color='k', linewidth=2, linestyle='--') 143 | index_last = index_last + 1 + 1 144 | plt.grid() 145 | plt.legend(('predicted', 'actual'), loc='upper right') 146 | plt.ylabel('Character value', FontSize=12) 147 | plt.xlabel('Word time slot', FontSize=12) 148 | plt.xticks(FontSize=12) 149 | plt.yticks(FontSize=12) 150 | 151 | plt.subplot2grid((2, 1), (1, 0)) 152 | index_last = 0 153 | for i in range(900, 910): 154 | sample_batched = [users_dataset_test[i]] 155 | sample_batched_sent = [sample_batched[i][0] for i in range(len(sample_batched))] 156 | if sample_batched_sent[0].shape[0] > 0: 157 | sample_batched_labels = [sample_batched[i][1] for i in range(len(sample_batched))] 158 | sample_batched_time = [sample_batched[i][2][1] for i in range(len(sample_batched))] 159 | sample_batched_duration = [sample_batched[i][2][0]*10 for i in range(len(sample_batched))] 160 | 161 | sample_batched_sent = torch.stack(sample_batched_sent) 162 | sample_batched_sent = torch.transpose(sample_batched_sent, 0, 1) 163 | x_input = sample_batched_sent 164 | x_input_array = x_input.data.numpy() 165 | x_input = torch.reshape(x_input, (x_input.shape[0], 1, 1)) 166 | hidden_layer = fully_connect(encoder_model(x_input)) 167 | representation = torch.transpose(hidden_layer, 0, 1) 168 | representation = representation.view(1, -1) 169 | representation_array = representation.data.numpy() 170 | 171 | dec_out = [torch.zeros(1, 1, 1)] 172 | out_vector = torch.zeros(sample_batched_sent.shape[0], 1) 173 | dec_hidden = hidden_layer[:, :, :] 174 | for word in range(out_vector.shape[0]): 175 | dec_out = input_packing(dec_out) 176 | out, dec_hidden = decoder_model(dec_out, dec_hidden) 177 | dec_out = out 178 | out_vector[word] = dec_out 179 | 180 | out_vector_array = pad_sequence(out_vector).data[0].numpy() 181 | x_input_array = x_input.data.numpy()[:, 0, 0] 182 | 183 | x_ax = np.linspace(index_last, 184 | out_vector_array.shape[0] + index_last, 185 | out_vector_array.shape[0]) 186 | index_last = index_last + out_vector_array.shape[0] 187 | 188 | mk = matplotlib.markers.MarkerStyle(marker='.', fillstyle='none') 189 | plt.plot(x_ax, out_vector_array*20000, c='firebrick', marker='x', linestyle='--', markersize=3, 190 | linewidth=0.8) 191 | plt.plot(x_ax, x_input_array*20000, c='midnightblue', marker='o', markersize=3, linewidth=0.8) 192 | plt.axvline(x=index_last + 1, color='k', linewidth=2, linestyle='--') 193 | index_last = index_last + 1 + 1 194 | plt.grid() 195 | plt.legend(('predicted', 'actual'), loc='upper right') 196 | plt.ylabel('Character value', FontSize=12) 197 | plt.xlabel('Word time slot', FontSize=12) 198 | plt.xticks(FontSize=12) 199 | plt.yticks(FontSize=12) 200 | 201 | fig.tight_layout() 202 | fig.set_size_inches(w=11, h=4) 203 | fig.savefig('../plots/autoencoder_multiple.eps') 204 | -------------------------------------------------------------------------------- /code/Bhattacharyya_distance.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | Battacharyya_distance: computes distance metrics 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import argparse 23 | import numpy as np 24 | import matplotlib 25 | import matplotlib.pyplot as plt 26 | import math as mt 27 | import pickle 28 | import matplotlib.gridspec as gridspec 29 | import scipy.io as sio 30 | matplotlib.use('TkAgg') 31 | plt.switch_backend('agg') 32 | 33 | if __name__ == '__main__': 34 | parser = argparse.ArgumentParser(description=__doc__) 35 | parser.add_argument('num_clusters', help='Number of classes for clustering', type=int) 36 | args = parser.parse_args() 37 | 38 | num_traces = 40 39 | 40 | num_clusters = args.num_clusters 41 | compute_pdf = True 42 | 43 | if compute_pdf: 44 | # CLUSTERING 45 | with open('../processed_files/gauss.txt', "rb") as fp: # Unpickling 46 | gauss = pickle.load(fp) 47 | 48 | # COMPUTE THE USERS PDF ON THE TRAINING DATA 49 | winLen = 3000 50 | winStep = 30 51 | users_pdf = np.zeros((num_traces, num_clusters)) 52 | 53 | for idx in range(num_traces): 54 | with open('../processed_files/hiddens_train_user' + str(idx) + '.txt', 55 | "rb") as fp: # Unpickling 56 | sentences = pickle.load(fp) 57 | with open('../processed_files/times_train_user' + str(idx) + '.txt', 58 | "rb") as fp: # Unpickling 59 | time = pickle.load(fp) 60 | 61 | time[:, 1] = time[:, 1] - time[0, 1] 62 | end = time[-1, 1] 63 | num_groups = int(mt.ceil((end - winLen) / winStep)) 64 | frequency_vector = np.zeros((num_groups, num_clusters)) 65 | for i in range(num_groups): 66 | new_start = min(list(time[:, 1]), key=lambda x: abs(x - i * winStep)) 67 | new_end = min(list(time[:, 1]), key=lambda x: abs(x - (new_start + winLen))) 68 | new_start_idx = np.argwhere(time[:, 1] == new_start)[0, 0] 69 | new_end_idx = np.argwhere(time[:, 1] == new_end)[0, 0] 70 | 71 | sentences_sampled_user = sentences[new_start_idx:new_end_idx, :] 72 | time_sampled = time[new_start_idx:new_end_idx, :] 73 | 74 | hidden_batched_vector = sentences_sampled_user.cpu().data.numpy() 75 | hidden_batched_vector = hidden_batched_vector 76 | labels_tot = gauss.predict(hidden_batched_vector) 77 | histog = np.histogram(labels_tot, bins=np.linspace(0, num_clusters, num_clusters + 1)) 78 | freq = histog[0] 79 | frequency_vector[i, :] = freq / np.amax(freq) 80 | frequency_vector_mean = np.mean(frequency_vector, axis=0) 81 | users_pdf[idx, :] = frequency_vector_mean / np.sum(frequency_vector_mean) 82 | 83 | with open('../outputs/users_pdf.txt', "wb") as fp: # Pickling 84 | pickle.dump(users_pdf, fp) 85 | 86 | # COMPUTE THE USERS PDF ON THE TEST DATA 87 | winLen = 3000 88 | winStep = 30 89 | users_pdf = np.zeros((num_traces, num_clusters)) 90 | 91 | for idx in range(num_traces): 92 | with open('../processed_files/hiddens_test_user' + str(idx) + '.txt', 93 | "rb") as fp: # Unpickling 94 | sentences = pickle.load(fp) 95 | with open('../processed_files/times_test_user' + str(idx) + '.txt', 96 | "rb") as fp: # Unpickling 97 | time = pickle.load(fp) 98 | 99 | time[:, 1] = time[:, 1] - time[0, 1] 100 | end = time[-1, 1] 101 | num_groups = int(mt.ceil((end - winLen) / winStep)) 102 | frequency_vector = np.zeros((num_groups, num_clusters)) 103 | for i in range(num_groups): 104 | new_start = min(list(time[:, 1]), key=lambda x: abs(x - i * winStep)) 105 | new_end = min(list(time[:, 1]), key=lambda x: abs(x - (new_start + winLen))) 106 | new_start_idx = np.argwhere(time[:, 1] == new_start)[0, 0] 107 | new_end_idx = np.argwhere(time[:, 1] == new_end)[0, 0] 108 | 109 | sentences_sampled_user = sentences[new_start_idx:new_end_idx, :] 110 | time_sampled = time[new_start_idx:new_end_idx, :] 111 | 112 | hidden_batched_vector = sentences_sampled_user.cpu().data.numpy() 113 | hidden_batched_vector = hidden_batched_vector 114 | labels_tot = gauss.predict(hidden_batched_vector) 115 | histog = np.histogram(labels_tot, bins=np.linspace(0, num_clusters, num_clusters + 1)) 116 | freq = histog[0] 117 | frequency_vector[i, :] = freq / np.amax(freq) 118 | frequency_vector_mean = np.mean(frequency_vector, axis=0) 119 | users_pdf[idx, :] = frequency_vector_mean / np.sum(frequency_vector_mean) 120 | 121 | with open('../outputs/users_pdf_test.txt', "wb") as fp: # Pickling 122 | pickle.dump(users_pdf, fp) 123 | 124 | # COMPUTE THE DISTANCE METRICS FOR THE TRAIN DATA 125 | with open('../outputs/users_pdf.txt', "rb") as fp: # Unpickling 126 | users_pdf = pickle.load(fp) 127 | 128 | BC = np.zeros((num_traces, num_traces)) 129 | for idx1 in range(num_traces): 130 | pdf_user1 = users_pdf[idx1, :] 131 | for idx2 in range(num_traces): 132 | if idx2 > idx1: 133 | pdf_user2 = users_pdf[idx2, :] 134 | BC[idx1, idx2] = -mt.log(np.sum(np.sqrt(np.multiply(pdf_user1, pdf_user2)))) 135 | 136 | # COMPUTE THE DISTANCE METRICS FOR THE TEST DATA 137 | with open('../outputs/users_pdf_test.txt', "rb") as fp: # Unpickling 138 | users_pdf_test = pickle.load(fp) 139 | 140 | BC_test = np.zeros((num_traces, num_traces)) 141 | for idx1 in range(num_traces): 142 | pdf_user1 = users_pdf_test[idx1, :] 143 | for idx2 in range(num_traces): 144 | if idx2 > idx1: 145 | pdf_user2 = users_pdf_test[idx2, :] 146 | BC_test[idx1, idx2] = -mt.log(np.sum(np.sqrt(np.multiply(pdf_user1, pdf_user2)))) 147 | 148 | with open('../outputs/traces_test_accuracy.txt', "rb") as fp: # Unpickling 149 | traces_test_accuracy = pickle.load(fp) 150 | 151 | BC_test_line = BC_test.reshape(-1) 152 | traces_test_accuracy_line = traces_test_accuracy.reshape(-1) 153 | 154 | mApp_structure = sio.loadmat('../input_files/mApp.mat') 155 | mApp = mApp_structure['mApp'] 156 | 157 | hamming_dist_app = np.zeros((num_traces, num_traces)) 158 | matrix_app = np.zeros((num_traces, num_traces), dtype=int) 159 | for idx1 in range(num_traces): 160 | app_user1 = mApp[idx1, :] 161 | for idx2 in range(num_traces): 162 | app_user2 = mApp[idx2, :] 163 | if idx2 > idx1: 164 | dist_coeff = np.sum(np.abs(app_user1 - app_user2)) / (mt.pow(app_user1.shape[0], 2)) 165 | hamming_dist_app[idx1, idx2] = dist_coeff 166 | matrix_app[idx1, idx2] = np.sum(app_user1 * app_user2) # app in common 167 | 168 | hamming_dist_app_line = hamming_dist_app.reshape(-1) 169 | 170 | fig = plt.figure(1) 171 | gs = gridspec.GridSpec(3, 3, wspace=0.38, hspace=0.68) 172 | ax1 = plt.subplot(gs[:, :-1]) 173 | ax2 = plt.subplot(gs[0, -1:]) 174 | ax3 = plt.subplot(gs[1, -1:]) 175 | ax4 = plt.subplot(gs[2, -1:]) 176 | ax1.scatter(hamming_dist_app_line / np.amax(hamming_dist_app_line), traces_test_accuracy_line, s=10, linewidths=0, 177 | c='firebrick', marker='x', label='Actual applications', linewidth=0.7) 178 | ax1.scatter(BC_test_line, traces_test_accuracy_line, s=10, linewidths=0, edgecolors='midnightblue', marker='o', 179 | label='Estimated applications', linewidth=0.4, facecolors='none') 180 | ax1.grid() 181 | ax1.tick_params(axis='both', which='major', labelsize=12) 182 | ax1.set_xlabel('Distance', FontSize=16) 183 | ax1.set_ylabel('Successful user disambiguation rate', FontSize=16) 184 | ax1.legend() 185 | 186 | idx = 3 187 | ax2.bar(np.linspace(1, num_clusters, num_clusters), users_pdf[idx, :], color='midnightblue') 188 | ax2.grid() 189 | ax2.set_title('User' + str(idx + 1), FontSize=16) 190 | ax2.set_xlabel('Clusters', FontSize=12) 191 | ax2.set_ylabel('Probability', FontSize=12) 192 | ax2.tick_params(axis='both', which='major', labelsize=12) 193 | ax2.set_xticks(np.linspace(1, num_clusters, num_clusters)) 194 | ax2.locator_params(axis='x', nbins=5) 195 | ax2.locator_params(axis='y', nbins=5) 196 | 197 | idx = 4 198 | ax3.bar(np.linspace(1, num_clusters, num_clusters), users_pdf[idx, :], color='midnightblue') 199 | ax3.grid() 200 | ax3.set_title('User' + str(idx + 1), FontSize=16) 201 | ax3.set_xlabel('Clusters', FontSize=12) 202 | ax3.set_ylabel('Probability', FontSize=12) 203 | ax2.tick_params(axis='both', which='major', labelsize=12) 204 | ax3.set_xticks(np.linspace(1, num_clusters, num_clusters)) 205 | ax3.locator_params(axis='x', nbins=5) 206 | ax3.locator_params(axis='y', nbins=5) 207 | 208 | idx = 11 209 | ax4.bar(np.linspace(1, num_clusters, num_clusters), users_pdf[idx, :], color='midnightblue') 210 | ax4.grid() 211 | ax4.set_title('User' + str(idx + 1), FontSize=16) 212 | ax4.set_xlabel('Clusters', FontSize=12) 213 | ax4.set_ylabel('Probability', FontSize=12) 214 | ax2.tick_params(axis='both', which='major', labelsize=12) 215 | ax4.set_xticks(np.linspace(1, num_clusters, num_clusters)) 216 | ax4.locator_params(axis='x', nbins=5) 217 | ax4.locator_params(axis='y', nbins=5) 218 | 219 | fig.tight_layout() 220 | fig.set_size_inches(w=11, h=7) 221 | fig.savefig('../plots/batt_pdf.eps') 222 | 223 | # Confusion Matrix with numbers of common applications 224 | confusion_matrix = np.load('../outputs/confusion_matrix_test.npy') 225 | 226 | number_users = confusion_matrix.shape[0] 227 | 228 | confusion_matrix_normaliz_row = confusion_matrix / np.sum(confusion_matrix, axis=1).reshape(-1, 1) 229 | confusion_matrix_normaliz_column = \ 230 | confusion_matrix_normaliz_row / np.sum(confusion_matrix_normaliz_row, axis=0).reshape(1, -1) 231 | max_columns = np.amax(confusion_matrix_normaliz_column, axis=0) 232 | sum_max_columns = np.sum(max_columns) 233 | 234 | correct_windows = np.sum(np.diag(confusion_matrix)) 235 | number_windows = np.sum(np.sum(confusion_matrix, 1)) 236 | perc_correct_window = correct_windows / number_windows * 100 237 | print('perc_correct_window: ' + str(perc_correct_window)) 238 | 239 | fig = plt.figure(6) 240 | ax = plt.axes() 241 | fig.set_size_inches(6, 5) 242 | max_matrix_app = np.amax(confusion_matrix_normaliz_row) 243 | im1 = plt.pcolor(np.linspace(0.5, num_traces + 0.5, num_traces + 1), 244 | np.linspace(0.5, num_traces + 0.5, num_traces + 1), 245 | np.transpose(confusion_matrix_normaliz_row), 246 | cmap='Blues', edgecolors='black', vmin=0, vmax=max_matrix_app) 247 | ax.set_title(r"$\bf{" + "Normalized" + "}$" + " " r"$\bf{" + "confusion" + "}$" + " " r"$\bf{" + "matrix" + "}$", 248 | FontSize=9) 249 | ax.set_xlabel('Actual user', FontSize=8) 250 | 251 | ax.set_xticks(np.linspace(1, num_traces, num_traces), minor=True) 252 | ax.set_yticks(np.linspace(1, num_traces, num_traces), minor=True) 253 | ax.set_ylabel('Predicted user', FontSize=8) 254 | ax.tick_params(axis="x", labelsize=8) 255 | ax.tick_params(axis="y", labelsize=8) 256 | 257 | for y_ax in range(matrix_app.shape[0]): 258 | for x_ax in range(matrix_app.shape[1]): 259 | col = 'k' 260 | if confusion_matrix_normaliz_row[x_ax, y_ax] > 0.6: # [x, y] because plot the transpose version 261 | col = 'w' 262 | ax.text(x_ax + 1, y_ax + 1, '%d' % matrix_app[y_ax, x_ax], horizontalalignment='center', 263 | verticalalignment='center', fontsize=4, color=col) 264 | 265 | cbar = fig.colorbar(im1) 266 | cbar.ax.set_ylabel('Accuracy', FontSize=8) 267 | cbar.ax.tick_params(axis="y", labelsize=7) 268 | 269 | fig.savefig('../plots/users_apps.eps') 270 | 271 | with open('../outputs/matrix_app.txt', "wb") as fp: # Pickling 272 | pickle.dump(matrix_app, fp) 273 | -------------------------------------------------------------------------------- /code/users_identification.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | users_identification: trains and validates the CNN for users identification 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import argparse 23 | import numpy as np 24 | import matplotlib.pyplot as plt 25 | import tensorflow as tf 26 | from tensorflow.python.framework import ops 27 | import math as mt 28 | import pickle 29 | import gc 30 | plt.switch_backend('agg') 31 | 32 | 33 | def convert_to_one_hot(y, c): 34 | y = np.eye(c)[(y - 1).reshape(-1)].T 35 | return y 36 | 37 | 38 | def create_placeholders(max_length_sequences, n_y): 39 | x = tf.placeholder(tf.float32, shape=(None, max_length_sequences, 1)) 40 | y = tf.placeholder(tf.float32, shape=(None, n_y)) 41 | return x, y 42 | 43 | 44 | def initialize_parameters(): 45 | tf.set_random_seed(1) 46 | w1 = tf.get_variable("W1", [10, 1, 5], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 47 | w2 = tf.get_variable("W2", [5, 5, 10], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 48 | w3 = tf.get_variable("W3", [5, 10, 20], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 49 | w4 = tf.get_variable("W4", [3, 20, 30], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 50 | parameters = {"w1": w1, "w2": w2, "w3": w3, "w4": w4} 51 | return parameters 52 | 53 | 54 | def cnn_encoder(x, parameters): 55 | w1 = parameters['w1'] 56 | w2 = parameters['w2'] 57 | w3 = parameters['w3'] 58 | w4 = parameters['w4'] 59 | 60 | z1 = tf.nn.conv1d(x, w1, stride=1, padding='SAME') 61 | a1 = tf.nn.relu(z1) 62 | d1 = tf.nn.dropout(a1, 0.8) 63 | 64 | z2 = tf.nn.conv1d(d1, w2, stride=1, padding='SAME') 65 | a2 = tf.nn.relu(z2) 66 | d2 = tf.nn.dropout(a2, 0.8) 67 | 68 | z3 = tf.nn.conv1d(d2, w3, stride=1, padding='SAME') 69 | a3 = tf.nn.relu(z3) 70 | d3 = tf.nn.dropout(a3, 0.8) 71 | 72 | z4 = tf.nn.conv1d(d3, w4, stride=1, padding='SAME') 73 | a4 = tf.nn.relu(z4) 74 | 75 | p6 = tf.contrib.layers.flatten(a4) 76 | p6 = tf.layers.dropout(p6, rate=0.2) 77 | 78 | z6 = tf.contrib.layers.fully_connected(p6, num_traces, activation_fn=None) 79 | return z6 80 | 81 | 82 | def compute_cost(z, y): 83 | cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=z, labels=y)) 84 | return cost 85 | 86 | 87 | def confusion_matrix_computation(num_users, test_labels, prediction_test_labels): 88 | confusion_matrix = np.zeros((num_users, num_users)) 89 | for i_act in range(1, num_users + 1): # actual user 90 | indices_act_i = np.argwhere(test_labels == i_act)[:, 0] 91 | for j_act in range(1, num_users + 1): # predicted user 92 | indices_act_j = np.argwhere(prediction_test_labels == j_act)[:, 0] 93 | intersect = set(indices_act_i).intersection(indices_act_j) 94 | confusion_matrix[i_act - 1, j_act - 1] = len(intersect) 95 | return confusion_matrix 96 | 97 | 98 | def random_mini_batches(x, y, mini_batch_size=64, seed=0): 99 | m = x.shape[0] # number of training examples 100 | mini_batches = [] 101 | np.random.seed(seed) 102 | 103 | permutation = list(np.random.permutation(m)) 104 | shuffled_x = x[permutation, :, :] 105 | shuffled_y = y[permutation, :] 106 | 107 | num_complete_minibatches = mt.floor( 108 | m / mini_batch_size) # number of mini batches of size mini_batch_size in your partitioning 109 | for k in range(0, num_complete_minibatches): 110 | mini_batch_x = shuffled_x[k * mini_batch_size: k * mini_batch_size + mini_batch_size, :, :] 111 | mini_batch_y = shuffled_y[k * mini_batch_size: k * mini_batch_size + mini_batch_size, :] 112 | mini_batch = (mini_batch_x, mini_batch_y) 113 | mini_batches.append(mini_batch) 114 | 115 | # Handling the end case (last mini-batch < mini_batch_size) 116 | if m % mini_batch_size != 0: 117 | mini_batch_x = shuffled_x[num_complete_minibatches * mini_batch_size: m, :, :] 118 | mini_batch_y = shuffled_y[num_complete_minibatches * mini_batch_size: m, :] 119 | mini_batch = (mini_batch_x, mini_batch_y) 120 | mini_batches.append(mini_batch) 121 | 122 | return mini_batches 123 | 124 | 125 | def model(x_train, y_train, x_test, y_test, test_accuracy_old, num_epochs=100, minibatch_size=64, print_cost=True): 126 | ops.reset_default_graph() # to be able to rerun the model without overwriting tf variables 127 | tf.set_random_seed(1) 128 | seed = 3 129 | (m, max_length_sequences, depth) = x_train.shape # m is the number of sequences 130 | (m_y, n_y) = y_train.shape 131 | 132 | costs = [] 133 | 134 | x, y = create_placeholders(max_length_sequences, n_y) 135 | 136 | parameters = initialize_parameters() 137 | 138 | z6 = cnn_encoder(x, parameters) 139 | 140 | cost = compute_cost(z6, y) 141 | 142 | optimizer = tf.train.AdamOptimizer().minimize(cost) 143 | 144 | init = tf.global_variables_initializer() 145 | 146 | train_accuracy = 0 147 | test_accuracy = 0 148 | 149 | # Start the session to compute the tensorflow graph 150 | with tf.Session() as sess: 151 | sess.run(init) 152 | 153 | for epoch in range(num_epochs): 154 | minibatch_cost = 0. 155 | num_minibatches = int(m / minibatch_size) # number of minibatches of size minibatch_size in the train set 156 | seed = seed + 1 157 | minibatches = random_mini_batches(x_train, y_train, minibatch_size, seed) 158 | 159 | for minibatch in minibatches: 160 | (minibatch_x, minibatch_y) = minibatch 161 | 162 | _, temp_cost = sess.run([optimizer, cost], feed_dict={x: minibatch_x, y: minibatch_y}) 163 | 164 | minibatch_cost += temp_cost / num_minibatches 165 | 166 | # Print the cost 167 | if print_cost and epoch % 5 == 0: 168 | print("Cost after epoch %i: %f" % (epoch, minibatch_cost)) 169 | 170 | predict_op = tf.argmax(z6, 1) 171 | correct_prediction = tf.equal(predict_op, tf.argmax(y, 1)) 172 | 173 | # Calculate accuracy on the test set 174 | accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float")) 175 | 176 | train_accuracy = accuracy.eval({x: x_train, y: y_train}) 177 | 178 | test_accuracy = accuracy.eval({x: x_test, y: y_test}) 179 | 180 | print("Train accuracy: %f" % train_accuracy) 181 | print("Test accuracy:%f " % test_accuracy) 182 | 183 | if print_cost and epoch % 10 == 0: 184 | costs.append(minibatch_cost) 185 | 186 | # Save the prediction 187 | prediction_train = sess.run(z6, {x: x_train, y: y_train}) 188 | prediction_test = sess.run(z6, {x: x_test, y: y_test}) 189 | 190 | prediction_train_label = np.argmax(prediction_train, axis=1) + 1 191 | prediction_test_label = np.argmax(prediction_test, axis=1) + 1 192 | 193 | if test_accuracy > test_accuracy_old: 194 | confusion_matrix_test = confusion_matrix_computation(num_traces, test_label, prediction_test_label) 195 | confusion_matrix_train = confusion_matrix_computation(num_traces, train_label, prediction_train_label) 196 | test_accuracy_old = test_accuracy 197 | 198 | gc.collect() 199 | np.save("../outputs/confusion_matrix_test.npy", confusion_matrix_test) 200 | np.save("../outputs/confusion_matrix_train.npy", confusion_matrix_train) 201 | 202 | sess.close() 203 | 204 | return train_accuracy, test_accuracy, parameters, prediction_train, prediction_test 205 | 206 | 207 | if __name__ == '__main__': 208 | parser = argparse.ArgumentParser(description=__doc__) 209 | parser.add_argument('num_clusters', help='Number of classes for clustering', type=int) 210 | parser.add_argument('epochs', help='Number of epochs', type=int) 211 | args = parser.parse_args() 212 | 213 | num_traces = 40 214 | 215 | # ---------------------------------------------------------------------------------------------------------------- # 216 | # CONVOLUTIONAL NETWORK USER IDENTIFICATION # 217 | 218 | num_clusters = args.num_clusters 219 | with open('../processed_files/gauss.txt', "rb") as fp: # Unpickling 220 | gauss = pickle.load(fp) 221 | 222 | num_epcs = args.epochs 223 | len_win = 3000 # in seconds 224 | step_win = 30 # in seconds 225 | users_sliding = [] 226 | index_vector = [] 227 | user_new = 0 228 | for user in range(num_traces): 229 | print(user) 230 | with open('../processed_files/hiddens_' + 'train' + '_user' + str(user) + '.txt', 231 | "rb") as fp: # Unpickling 232 | sentences_u = pickle.load(fp) 233 | with open('../processed_files/times_' + 'train' + '_user' + str(user) + '.txt', 234 | "rb") as fp: # Unpickling 235 | time_u = pickle.load(fp) 236 | sentences_u = sentences_u.data.cpu().numpy() 237 | time_ass = time_u[:, 1] - time_u[0, 1] 238 | length_user = time_ass[-1] 239 | sentences_k = [] 240 | num_window = mt.ceil((length_user-len_win)/step_win) 241 | 242 | print('user ' + str(user) + ': train window ' + str(num_window)) 243 | for t in range(num_window): 244 | indices = np.argwhere((time_ass > t*step_win) & (time_ass < t*step_win + len_win))[:, 0] 245 | if indices.shape[0] > 1: 246 | sentences_k.append(sentences_u[indices[0]:indices[-1], :]) 247 | 248 | frequency_vector = np.zeros((len(sentences_k), num_clusters)) 249 | for s in range(len(sentences_k)): 250 | sentence = np.asarray(sentences_k[s]) 251 | hidden_batched_vector = sentence 252 | labels_tot = gauss.predict(hidden_batched_vector) 253 | histog = np.histogram(labels_tot, bins=np.linspace(0, num_clusters, num_clusters + 1)) 254 | freq = histog[0] 255 | frequency_vector[s, :] = freq/np.amax(freq) 256 | 257 | users_sliding.append(frequency_vector) 258 | index_vector.extend(len(sentences_k) * [user_new]) 259 | user_new = user_new + 1 260 | 261 | input_matrix_tr = np.vstack(users_sliding) 262 | train_label = np.asarray(index_vector) + 1 263 | 264 | users_sliding = [] 265 | index_vector = [] 266 | len_win = 3000 # in seconds 267 | step_win = 30 # in seconds 268 | user_new = 0 269 | for user in range(num_traces): 270 | print(user) 271 | with open('../processed_files/hiddens_test_user' + str(user) + '.txt', 272 | "rb") as fp: # Unpickling 273 | sentences_u = pickle.load(fp) 274 | with open('../processed_files/times_test_user' + str(user) + '.txt', 275 | "rb") as fp: # Unpickling 276 | time_u = pickle.load(fp) 277 | sentences_u = sentences_u.data.cpu().numpy() 278 | time_ass = time_u[:, 1] - time_u[0, 1] 279 | length_user = time_ass[-1] 280 | sentences_k = [] 281 | num_window = mt.ceil((length_user-len_win)/step_win) 282 | print('user ' + str(user) + ': test window ' + str(num_window)) 283 | for t in range(num_window): 284 | indices = np.argwhere((time_ass > t*step_win) & (time_ass < t*step_win + len_win))[:, 0] 285 | if indices.shape[0] > 1: 286 | sentences_k.append(sentences_u[indices[0]:indices[-1], :]) 287 | 288 | frequency_vector = np.zeros((len(sentences_k), num_clusters)) 289 | for s in range(len(sentences_k)): 290 | sentence = np.asarray(sentences_k[s]) 291 | hidden_batched_vector = sentence 292 | labels_tot = gauss.predict(hidden_batched_vector) 293 | histog = np.histogram(labels_tot, bins=np.linspace(0, num_clusters, num_clusters + 1)) 294 | freq = histog[0] 295 | frequency_vector[s, :] = freq/np.amax(freq) 296 | 297 | users_sliding.append(frequency_vector) 298 | index_vector.extend(len(sentences_k) * [user_new]) 299 | user_new = user_new + 1 300 | 301 | input_matrix_test = np.vstack(users_sliding) 302 | test_label = np.asarray(index_vector) + 1 303 | 304 | input_matrix_tr = input_matrix_tr.reshape((input_matrix_tr.shape[0], input_matrix_tr.shape[1], 1)) 305 | input_matrix_test = input_matrix_test.reshape((input_matrix_test.shape[0], input_matrix_test.shape[1], 1)) 306 | 307 | output_matrix_tr = convert_to_one_hot(train_label, num_traces).T 308 | output_matrix_test = convert_to_one_hot(test_label, num_traces).T 309 | 310 | test_accuracy_input = 0 311 | _, test_accuracy_new, parameters_out, prediction_tr, prediction_tst = model(input_matrix_tr, output_matrix_tr, 312 | input_matrix_test, output_matrix_test, 313 | test_accuracy_input, 314 | num_epochs=num_epcs) 315 | -------------------------------------------------------------------------------- /code/users_disambiguation.py: -------------------------------------------------------------------------------- 1 | 2 | """ 3 | users_disambiguation: computes the accuracy in the separation between two users 4 | 5 | Copyright (C) 2019 Francesca Meneghello, Michele Rossi, Nicola Bui 6 | contact: meneghello@dei.unipd.it 7 | 8 | This program is free software: you can redistribute it and/or modify 9 | it under the terms of the GNU General Public License as published by 10 | the Free Software Foundation, either version 3 of the License, or 11 | (at your option) any later version. 12 | 13 | This program is distributed in the hope that it will be useful, 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 | GNU General Public License for more details. 17 | 18 | You should have received a copy of the GNU General Public License 19 | along with this program. If not, see . 20 | """ 21 | 22 | import argparse 23 | import numpy as np 24 | import matplotlib.pyplot as plt 25 | import tensorflow as tf 26 | from tensorflow.python.framework import ops 27 | import math as mt 28 | import pickle 29 | import gc 30 | from pathlib import Path 31 | plt.switch_backend('agg') 32 | 33 | 34 | def convert_to_one_hot(y, c): 35 | y = np.eye(c)[(y - 1).reshape(-1)].T 36 | return y 37 | 38 | 39 | def create_placeholders(max_length_sequences, n_y): 40 | x = tf.placeholder(tf.float32, shape=(None, max_length_sequences, 1)) 41 | y = tf.placeholder(tf.float32, shape=(None, n_y)) 42 | return x, y 43 | 44 | 45 | def initialize_parameters(): 46 | tf.set_random_seed(1) 47 | w1 = tf.get_variable("W1", [10, 1, 5], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 48 | w2 = tf.get_variable("W2", [5, 5, 10], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 49 | w3 = tf.get_variable("W3", [5, 10, 20], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 50 | w4 = tf.get_variable("W4", [3, 20, 30], initializer=tf.contrib.layers.xavier_initializer(seed=0)) 51 | parameters = {"w1": w1, "w2": w2, "w3": w3, "w4": w4} 52 | return parameters 53 | 54 | 55 | def cnn_encoder(x, parameters): 56 | w1 = parameters['w1'] 57 | w2 = parameters['w2'] 58 | w3 = parameters['w3'] 59 | w4 = parameters['w4'] 60 | 61 | z1 = tf.nn.conv1d(x, w1, stride=1, padding='SAME') 62 | a1 = tf.nn.relu(z1) 63 | d1 = tf.nn.dropout(a1, 0.8) 64 | 65 | z2 = tf.nn.conv1d(d1, w2, stride=1, padding='SAME') 66 | a2 = tf.nn.relu(z2) 67 | d2 = tf.nn.dropout(a2, 0.8) 68 | 69 | z3 = tf.nn.conv1d(d2, w3, stride=1, padding='SAME') 70 | a3 = tf.nn.relu(z3) 71 | d3 = tf.nn.dropout(a3, 0.8) 72 | 73 | z4 = tf.nn.conv1d(d3, w4, stride=1, padding='SAME') 74 | a4 = tf.nn.relu(z4) 75 | 76 | p6 = tf.contrib.layers.flatten(a4) 77 | p6 = tf.layers.dropout(p6, rate=0.2) 78 | 79 | z6 = tf.contrib.layers.fully_connected(p6, num_selected_traces, activation_fn=None) 80 | return z6 81 | 82 | 83 | def compute_cost(z, y): 84 | cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=z, labels=y)) 85 | return cost 86 | 87 | 88 | def confusion_matrix_computation(num_users, test_labels, prediction_test_labels): 89 | confusion_matrix = np.zeros((num_users, num_users)) 90 | for i_act in range(1, num_users + 1): # actual user 91 | indices_act_i = np.argwhere(test_labels == i_act)[:, 0] 92 | for j_act in range(1, num_users + 1): # predicted user 93 | indices_act_j = np.argwhere(prediction_test_labels == j_act)[:, 0] 94 | intersect = set(indices_act_i).intersection(indices_act_j) 95 | confusion_matrix[i_act - 1, j_act - 1] = len(intersect) 96 | return confusion_matrix 97 | 98 | 99 | def random_mini_batches(x, y, mini_batch_size=64, seed=0): 100 | m = x.shape[0] # number of training examples 101 | mini_batches = [] 102 | np.random.seed(seed) 103 | 104 | permutation = list(np.random.permutation(m)) 105 | shuffled_x = x[permutation, :, :] 106 | shuffled_y = y[permutation, :] 107 | 108 | num_complete_minibatches = mt.floor( 109 | m / mini_batch_size) # number of mini batches of size mini_batch_size in your partitioning 110 | for k in range(0, num_complete_minibatches): 111 | mini_batch_x = shuffled_x[k * mini_batch_size: k * mini_batch_size + mini_batch_size, :, :] 112 | mini_batch_y = shuffled_y[k * mini_batch_size: k * mini_batch_size + mini_batch_size, :] 113 | mini_batch = (mini_batch_x, mini_batch_y) 114 | mini_batches.append(mini_batch) 115 | 116 | # Handling the end case (last mini-batch < mini_batch_size) 117 | if m % mini_batch_size != 0: 118 | mini_batch_x = shuffled_x[num_complete_minibatches * mini_batch_size: m, :, :] 119 | mini_batch_y = shuffled_y[num_complete_minibatches * mini_batch_size: m, :] 120 | mini_batch = (mini_batch_x, mini_batch_y) 121 | mini_batches.append(mini_batch) 122 | 123 | return mini_batches 124 | 125 | 126 | def model(x_train, y_train, x_test, y_test, test_accuracy_old, num_epochs=100, minibatch_size=64, print_cost=True): 127 | ops.reset_default_graph() # to be able to rerun the model without overwriting tf variables 128 | tf.set_random_seed(1) 129 | seed = 3 130 | (m, max_length_sequences, depth) = x_train.shape # m is the number of sequences 131 | (m_y, n_y) = y_train.shape 132 | 133 | costs = [] 134 | 135 | x, y = create_placeholders(max_length_sequences, n_y) 136 | 137 | parameters = initialize_parameters() 138 | 139 | z6 = cnn_encoder(x, parameters) 140 | 141 | cost = compute_cost(z6, y) 142 | 143 | optimizer = tf.train.AdamOptimizer().minimize(cost) 144 | 145 | init = tf.global_variables_initializer() 146 | 147 | train_accuracy = 0 148 | test_accuracy = 0 149 | 150 | # Start the session to compute the tensorflow graph 151 | with tf.Session() as sess: 152 | sess.run(init) 153 | 154 | for epoch in range(num_epochs): 155 | minibatch_cost = 0. 156 | num_minibatches = int(m / minibatch_size) # number of minibatches of size minibatch_size in the train set 157 | seed = seed + 1 158 | minibatches = random_mini_batches(x_train, y_train, minibatch_size, seed) 159 | 160 | for minibatch in minibatches: 161 | (minibatch_x, minibatch_y) = minibatch 162 | 163 | _, temp_cost = sess.run([optimizer, cost], feed_dict={x: minibatch_x, y: minibatch_y}) 164 | 165 | minibatch_cost += temp_cost / num_minibatches 166 | 167 | # Print the cost 168 | if print_cost and epoch % 5 == 0: 169 | print("Cost after epoch %i: %f" % (epoch, minibatch_cost)) 170 | 171 | predict_op = tf.argmax(z6, 1) 172 | correct_prediction = tf.equal(predict_op, tf.argmax(y, 1)) 173 | 174 | # Calculate accuracy on the test set 175 | accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float")) 176 | 177 | train_accuracy = accuracy.eval({x: x_train, y: y_train}) 178 | 179 | test_accuracy = accuracy.eval({x: x_test, y: y_test}) 180 | 181 | print("Train accuracy: %f" % train_accuracy) 182 | print("Test accuracy:%f " % test_accuracy) 183 | 184 | if print_cost and epoch % 10 == 0: 185 | costs.append(minibatch_cost) 186 | 187 | # Save the prediction 188 | prediction_train = sess.run(z6, {x: x_train, y: y_train}) 189 | prediction_test = sess.run(z6, {x: x_test, y: y_test}) 190 | 191 | if test_accuracy > test_accuracy_old: 192 | test_accuracy_old = test_accuracy 193 | 194 | gc.collect() 195 | 196 | sess.close() 197 | 198 | return train_accuracy, test_accuracy, parameters, prediction_train, prediction_test 199 | 200 | 201 | if __name__ == '__main__': 202 | parser = argparse.ArgumentParser(description=__doc__) 203 | parser.add_argument('num_clusters', help='Number of classes for clustering', type=int) 204 | parser.add_argument('epochs', help='Number of epochs', type=int) 205 | args = parser.parse_args() 206 | 207 | num_traces = 40 208 | 209 | # ---------------------------------------------------------------------------------------------------------------- # 210 | # CONVOLUTIONAL NETWORK USER IDENTIFICATION # 211 | 212 | num_clusters = args.num_clusters 213 | with open('../processed_files/gauss.txt', "rb") as fp: # Unpickling 214 | gauss = pickle.load(fp) 215 | 216 | num_selected_traces = 2 217 | 218 | num_epcs = args.epochs 219 | 220 | my_file = Path('../outputs/traces_test_accuracy.txt') 221 | if my_file.is_file(): 222 | with open('../outputs/traces_test_accuracy.txt', "rb") as fp: # Unpickling 223 | traces_test_accuracy = pickle.load(fp) 224 | start_idx1 = np.argwhere(np.diag(traces_test_accuracy, 1) == 0)[0, 0] - 1 225 | start_idx2 = np.argwhere(traces_test_accuracy[start_idx1, start_idx1 + 1:] == 0) 226 | if len(start_idx2) == 0: 227 | start_idx1 = start_idx1 + 1 228 | start_idx2 = start_idx1 + 1 229 | else: 230 | start_idx2 = start_idx2[0, 0] + start_idx1 + 1 231 | else: 232 | traces_test_accuracy = np.zeros((num_traces, num_traces)) 233 | start_idx1 = 0 234 | start_idx2 = 1 235 | 236 | for idx1 in range(start_idx1, num_traces): 237 | if start_idx1 != idx1: 238 | start_idx2 = idx1 + 1 239 | for idx2 in range(start_idx2, num_traces): 240 | print(' ') 241 | print('idx1 ' + str(idx1)) 242 | print('idx2 ' + str(idx2)) 243 | selected_indices = np.asarray([idx1, idx2]) 244 | 245 | len_win = 3000 # in seconds 246 | step_win = 30 # in seconds 247 | users_sliding = [] 248 | index_vector = [] 249 | user_new = 0 250 | for user in selected_indices: 251 | print(user) 252 | with open('../processed_files/hiddens_train_user' + str(user) + '.txt', 253 | "rb") as fp: # Unpickling 254 | sentences_u = pickle.load(fp) 255 | with open('../processed_files/times_train_user' + str(user) + '.txt', 256 | "rb") as fp: # Unpickling 257 | time_u = pickle.load(fp) 258 | sentences_u = sentences_u.data.cpu().numpy() 259 | time_ass = time_u[:, 1] - time_u[0, 1] 260 | length_user = time_ass[-1] 261 | sentences_k = [] 262 | num_window = mt.ceil((length_user - len_win) / step_win) 263 | 264 | print('user ' + str(user) + ': train window ' + str(num_window)) 265 | for t in range(num_window): 266 | indices = np.argwhere((time_ass > t * step_win) & (time_ass < t * step_win + len_win))[:, 0] 267 | if indices.shape[0] > 1: 268 | sentences_k.append(sentences_u[indices[0]:indices[-1], :]) 269 | 270 | frequency_vector = np.zeros((len(sentences_k), num_clusters)) 271 | for s in range(len(sentences_k)): 272 | sentence = np.asarray(sentences_k[s]) 273 | hidden_batched_vector = sentence 274 | labels_tot = gauss.predict(hidden_batched_vector) 275 | histog = np.histogram(labels_tot, bins=np.linspace(0, num_clusters, num_clusters + 1)) 276 | freq = histog[0] 277 | frequency_vector[s, :] = freq / np.amax(freq) 278 | 279 | users_sliding.append(frequency_vector) 280 | index_vector.extend(len(sentences_k) * [user_new]) 281 | user_new = user_new + 1 282 | 283 | input_matrix_tr = np.vstack(users_sliding) 284 | train_label = np.asarray(index_vector) + 1 285 | 286 | users_sliding = [] 287 | index_vector = [] 288 | len_win = 3000 # in seconds 289 | step_win = 30 # in seconds 290 | user_new = 0 291 | for user in selected_indices: 292 | print(user) 293 | with open('../processed_files/hiddens_test_user' + str(user) + '.txt', 294 | "rb") as fp: # Unpickling 295 | sentences_u = pickle.load(fp) 296 | with open('../processed_files/times_test_user' + str(user) + '.txt', 297 | "rb") as fp: # Unpickling 298 | time_u = pickle.load(fp) 299 | sentences_u = sentences_u.data.cpu().numpy() 300 | time_ass = time_u[:, 1] - time_u[0, 1] 301 | length_user = time_ass[-1] 302 | sentences_k = [] 303 | num_window = mt.ceil((length_user - len_win) / step_win) 304 | 305 | print('user ' + str(user) + ': test window ' + str(num_window)) 306 | for t in range(num_window): 307 | indices = np.argwhere((time_ass > t * step_win) & (time_ass < t * step_win + len_win))[:, 0] 308 | if indices.shape[0] > 1: 309 | sentences_k.append(sentences_u[indices[0]:indices[-1], :]) 310 | 311 | frequency_vector = np.zeros((len(sentences_k), num_clusters)) 312 | for s in range(len(sentences_k)): 313 | sentence = np.asarray(sentences_k[s]) 314 | hidden_batched_vector = sentence 315 | labels_tot = gauss.predict(hidden_batched_vector) 316 | histog = np.histogram(labels_tot, bins=np.linspace(0, num_clusters, num_clusters + 1)) 317 | freq = histog[0] 318 | frequency_vector[s, :] = freq/np.amax(freq) 319 | 320 | users_sliding.append(frequency_vector) 321 | index_vector.extend(len(sentences_k) * [user_new]) 322 | user_new = user_new + 1 323 | 324 | input_matrix_test = np.vstack(users_sliding) 325 | test_label = np.asarray(index_vector) + 1 326 | 327 | input_matrix_tr = input_matrix_tr.reshape((input_matrix_tr.shape[0], input_matrix_tr.shape[1], 1)) 328 | input_matrix_test = input_matrix_test.reshape((input_matrix_test.shape[0], input_matrix_test.shape[1], 1)) 329 | 330 | output_matrix_tr = convert_to_one_hot(train_label, num_selected_traces).T 331 | output_matrix_test = convert_to_one_hot(test_label, num_selected_traces).T 332 | 333 | test_acc_old = 0 334 | _, test_accuracy_new, _, _, _ = model(input_matrix_tr, output_matrix_tr, input_matrix_test, 335 | output_matrix_test, test_acc_old, num_epochs=num_epcs) 336 | 337 | traces_test_accuracy[idx1, idx2] = test_accuracy_new 338 | 339 | with open('../processed_files/traces_test_accuracy.txt', "wb") as fp: # Pickling 340 | pickle.dump(traces_test_accuracy, fp) 341 | del input_matrix_tr 342 | del input_matrix_test 343 | del output_matrix_test 344 | del output_matrix_tr 345 | del test_label 346 | gc.collect() 347 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | --------------------------------------------------------------------------------