updated model to use dense classifier

speech_data.py

@@ -107,8 +107,8 @@ def create_speech_pairs_data(audio_group='audio'):
 def speech_model_data():
     tr_pairs = np.load('outputs/tr_pairs.npy') / 255.0
     te_pairs = np.load('outputs/te_pairs.npy') / 255.0
-    # tr_pairs[tr_pairs < 0] = 0
-    # te_pairs[te_pairs < 0] = 0
+    tr_pairs[tr_pairs < 0] = 0
+    te_pairs[te_pairs < 0] = 0
     tr_y = np.load('outputs/tr_y.npy')
     te_y = np.load('outputs/te_y.npy')
     return tr_pairs, te_pairs, tr_y, te_y

speech_siamese.py

@@ -3,7 +3,10 @@ from __future__ import print_function
 import numpy as np
 from speech_data import speech_model_data
 from keras.models import Model,load_model
-from keras.layers import Input, Dense, Dropout, LSTM, Lambda
+from keras.layers import Input, Dense, Dropout, LSTM, Lambda, Concatenate
+# from keras.losses import categorical_crossentropy
+from keras.losses import binary_crossentropy
+# from keras.utils.np_utils import to_categorical
 from keras.optimizers import RMSprop
 from keras.callbacks import TensorBoard, ModelCheckpoint
 from keras import backend as K
@@ -34,20 +37,9 @@ def create_base_rnn_network(input_dim):
     inp = Input(shape=input_dim)
     ls1 = LSTM(1024, return_sequences=True)(inp)
     ls2 = LSTM(512, return_sequences=True)(ls1)
-    ls3 = LSTM(32)(ls2)
-    return Model(inp, ls3)
-
-
-def create_base_network(input_dim):
-    '''Base network to be shared (eq. to feature extraction).
-    '''
-    input = Input(shape=input_dim)
-    x = Dense(128, activation='relu')(input)
-    x = Dropout(0.1)(x)
-    x = Dense(128, activation='relu')(x)
-    x = Dropout(0.1)(x)
-    x = Dense(128, activation='relu')(x)
-    return Model(input, x)
+    # ls3 = LSTM(32, return_sequences=True)(ls2)
+    ls4 = LSTM(32)(ls2)
+    return Model(inp, ls4)
 
 
 def compute_accuracy(y_true, y_pred):
@@ -62,6 +54,13 @@ def accuracy(y_true, y_pred):
     '''
     return K.mean(K.equal(y_true, K.cast(y_pred < 0.5, y_true.dtype)))
 
+def dense_classifier(processed):
+    conc_proc = Concatenate()(processed)
+    d1 = Dense(8, activation='relu')(conc_proc)
+    dr1 = Dropout(0.1)(d1)
+    # d2 = Dense(8, activation='relu')(dr1)
+    # dr2 = Dropout(0.1)(d2)
+    return Dense(1, activation='sigmoid')(dr1)
 
 def siamese_model(input_dim):
     # input_dim = (15, 1654)
@@ -70,11 +69,12 @@ def siamese_model(input_dim):
     input_b = Input(shape=input_dim)
     processed_a = base_network(input_a)
     processed_b = base_network(input_b)
-    distance = Lambda(
-        euclidean_distance,
-        output_shape=eucl_dist_output_shape)([processed_a, processed_b])
-
-    model = Model([input_a, input_b], distance)
+    final_output = dense_classifier([processed_a,processed_b])
+    model = Model([input_a, input_b], final_output)
+    # distance = Lambda(
+    #     euclidean_distance,
+    #     output_shape=eucl_dist_output_shape)([processed_a, processed_b])
+    # model = Model([input_a, input_b], distance)
     return model
 
 
@@ -95,12 +95,12 @@ def train_siamese():
         embeddings_freq=0,
         embeddings_layer_names=None,
         embeddings_metadata=None)
-    cp_file_fmt = './models/siamese_speech_model-{epoch:02d}-epoch-{val_acc:0.2f}\
+    cp_file_fmt = './models/siamese_speech_model-{epoch:02d}-epoch-{val_loss:0.2f}\
     -acc.h5'
 
     cp_cb = ModelCheckpoint(
         cp_file_fmt,
-        monitor='val_acc',
+        monitor='val_loss',
         verbose=0,
         save_best_only=False,
         save_weights_only=False,
@@ -108,7 +108,7 @@ def train_siamese():
         period=1)
     # train
     rms = RMSprop(lr=0.001)
-    model.compile(loss=contrastive_loss, optimizer=rms, metrics=[accuracy])
+    model.compile(loss=binary_crossentropy, optimizer=rms, metrics=[accuracy])
     model.fit(
         [tr_pairs[:, 0], tr_pairs[:, 1]],
         tr_y,