updated test code

speech_siamese.py

@@ -4,8 +4,9 @@ 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, Concatenate
-# from keras.losses import categorical_crossentropy
-from keras.losses import binary_crossentropy
+from keras.losses import categorical_crossentropy
+# from keras.losses import binary_crossentropy
+from keras.utils import to_categorical
 # from keras.utils.np_utils import to_categorical
 from keras.optimizers import RMSprop
 from keras.callbacks import TensorBoard, ModelCheckpoint
@@ -30,15 +31,14 @@ def contrastive_loss(y_true, y_pred):
     return K.mean(y_true * K.square(y_pred) +
                   (1 - y_true) * K.square(K.maximum(1 - y_pred, 0)))
 
-
 def create_base_rnn_network(input_dim):
     '''Base network to be shared (eq. to feature extraction).
     '''
     inp = Input(shape=input_dim)
-    ls1 = LSTM(1024, return_sequences=True)(inp)
-    ls2 = LSTM(512, return_sequences=True)(ls1)
+    ls1 = LSTM(256, return_sequences=True)(inp)
+    ls2 = LSTM(128, return_sequences=True)(ls1)
     # ls3 = LSTM(32, return_sequences=True)(ls2)
-    ls4 = LSTM(32)(ls2)
+    ls4 = LSTM(64)(ls2)
     return Model(inp, ls4)
 
 
@@ -52,15 +52,15 @@ def compute_accuracy(y_true, y_pred):
 def accuracy(y_true, y_pred):
     '''Compute classification accuracy with a fixed threshold on distances.
     '''
-    return K.mean(K.equal(y_true, K.cast(y_pred < 0.5, y_true.dtype)))
+    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)
+    d1 = Dense(16, activation='relu')(conc_proc)
+    # dr1 = Dropout(0.1)(d1)
+    d2 = Dense(8, activation='relu')(d1)
     # dr2 = Dropout(0.1)(d2)
-    return Dense(1, activation='sigmoid')(dr1)
+    return Dense(2, activation='softmax')(d2)
 
 def siamese_model(input_dim):
     # input_dim = (15, 1654)
@@ -80,7 +80,9 @@ def siamese_model(input_dim):
 
 def train_siamese():
     # the data, shuffled and split between train and test sets
-    tr_pairs, te_pairs, tr_y, te_y = speech_model_data()
+    tr_pairs, te_pairs, tr_y_e, te_y_e = speech_model_data()
+    tr_y = to_categorical(tr_y_e, num_classes=2)
+    te_y = to_categorical(te_y_e, num_classes=2)
     input_dim = (tr_pairs.shape[2], tr_pairs.shape[3])
 
     model = siamese_model(input_dim)
@@ -96,7 +98,7 @@ def train_siamese():
         embeddings_layer_names=None,
         embeddings_metadata=None)
     cp_file_fmt = './models/siamese_speech_model-{epoch:02d}-epoch-{val_loss:0.2f}\
-    -acc.h5'
+-acc.h5'
 
     cp_cb = ModelCheckpoint(
         cp_file_fmt,
@@ -108,7 +110,7 @@ def train_siamese():
         period=1)
     # train
     rms = RMSprop(lr=0.001)
-    model.compile(loss=binary_crossentropy, optimizer=rms, metrics=[accuracy])
+    model.compile(loss=categorical_crossentropy, optimizer=rms, metrics=[accuracy])
     model.fit(
         [tr_pairs[:, 0], tr_pairs[:, 1]],
         tr_y,

test_siamese.py

@@ -5,6 +5,7 @@ import speech_data
 reload(speech_data)
 from speech_data import create_test_pair,get_word_pairs_data
 import numpy as np
+from keras.utils import to_categorical
 
 model = siamese_model((15, 1654))
 model.load_weights('./models/siamese_speech_model-final.h5')
@@ -17,6 +18,6 @@ def predict_recording_with(m,sample_size=15):
 
 predict_recording_with(model)
 
-sunflower_data,sunflower_result = get_word_pairs_data('sunflowers',15)
-sunflower_result
-model.predict([sunflower_data[:, 0], sunflower_data[:, 1]]) < 0.5
+sunflower_data,sunflower_result = get_word_pairs_data('sweater',15)
+print(np.argmax(model.predict([sunflower_data[:, 0], sunflower_data[:, 1]]),axis=1))
+print(sunflower_result)