Created
March 20, 2017 11:37
-
-
Save slashvar/286cfbeed568a70d1f29e4fa77d467b4 to your computer and use it in GitHub Desktop.
LSTM siamese network (masking issues)
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| from keras import backend as K | |
| from keras.layers import Input, Dense, merge, Dropout, Lambda, LSTM, Masking | |
| from keras.models import Model, Sequential | |
| from keras.optimizers import SGD, RMSprop, Adam, Nadam | |
| from sys import argv | |
| import argparse | |
| import csv | |
| import json | |
| import numpy as np | |
| import pickle | |
| import random | |
| DROPFACTOR=0.3 | |
| def core_layers(bucket_size, out_dims): | |
| seq = Sequential() | |
| seq.add(Masking(input_shape = (None, bucket_size), mask_value=0.0)) | |
| seq.add(LSTM(out_dims * 8, return_sequences=False, | |
| # input_shape = (None, bucket_size), | |
| kernel_initializer='he_normal', | |
| dropout=DROPFACTOR, recurrent_dropout=DROPFACTOR, implementation=1)) | |
| seq.add(Dropout(DROPFACTOR)) | |
| seq.add(Dense(out_dims * 8, activation='relu', kernel_initializer='he_normal')) | |
| seq.add(Dropout(DROPFACTOR)) | |
| seq.add(Dense(out_dims * 4, activation='relu', kernel_initializer='he_normal')) | |
| seq.add(Dropout(DROPFACTOR)) | |
| seq.add(Dense(out_dims * 2, activation='relu', kernel_initializer='he_normal')) | |
| seq.add(Dropout(DROPFACTOR)) | |
| seq.add(Dense(out_dims, activation='linear', kernel_initializer='he_normal')) | |
| return seq | |
| def euclidean_dist(vects): | |
| x, y = vects | |
| return K.sqrt(K.sum(K.square(x - y), axis=1, keepdims=True)) | |
| def eucl_dist_output_shape(shapes): | |
| shape1, shape2 = shapes | |
| return (shape1[0], 1) | |
| def contrastive_loss(y_true, y_pred): | |
| '''Contrastive loss from Hadsell-et-al.'06 | |
| http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf | |
| ''' | |
| margin = 1 | |
| return K.mean(y_true * K.square(y_pred) | |
| + (1 - y_true) * K.square(K.maximum(margin - y_pred, 0))) | |
| def build_model(bucket_size, out_dims): | |
| input1 = Input(shape=(None, bucket_size)) | |
| input2 = Input(shape=(None, bucket_size)) | |
| core = core_layers(bucket_size, out_dims) | |
| redux1 = core(input1) | |
| redux2 = core(input2) | |
| distance = Lambda(euclidean_dist, | |
| output_shape=eucl_dist_output_shape)([redux1, redux2]) | |
| model = Model(inputs=[input1, input2], outputs=distance) | |
| rms = RMSprop() | |
| model.compile(loss=contrastive_loss, optimizer=rms, metrics=['accuracy']) | |
| return model, core |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment