D = [27, 72, 70, 74, 57, 72, 19, 30, 75, 57]
for NConv, NJanet, NDense in zip(C, J, D):
acc_list = []
prec_list = []
recall_list = []
f1_list = []
support_list = []
for i in range(repetition):
early_stopping = EarlyStopping(monitor='val_acc', patience=patience)
checkpoint = ModelCheckpoint('6dmg_try.h5', monitor='val_acc', verbose=0, save_weights_only=True)
nadam = Nadam(lr=1e-4, beta_1=0.9, beta_2=0.9)
model1 = Sequential()
model1.add(Conv1D(NConv, 5, activation='relu', input_shape=(240, 6)))
model1.add(JANET(NJanet, activation='relu')) #, input_shape=(None, 6)))
model1.add(Dense(NDense, activation='relu'))
model1.add(Dense(20, activation='softmax'))
print model1.summary(90)
model1.compile(loss='categorical_crossentropy', optimizer=nadam, metrics=['accuracy'])
model1.fit(xTrain,
yTrain,
epochs=epochs,
validation_data=(xVal, yVal),
callbacks=[early_stopping],
verbose=1,
batch_size=batch_size)
predictedTest = np.argmax(model1.predict(xTest), axis=1)
acc = accuracy_score(yTest, predictedTest)
precision, recall, f1, support = precision_recall_fscore_support(yTest, predictedTest, average='weighted')
prec_list = []
recall_list = []
f1_list = []
for i in range(repetition):
early_stopping = EarlyStopping(monitor='val_acc', patience=patience)
checkpoint = ModelCheckpoint('6dmg_try.h5',
monitor='val_acc',
verbose=0,
save_weights_only=True)
nadam = Nadam(lr=1e-4, beta_1=0.9, beta_2=0.9)
model6 = Sequential()
model6.add(Conv1D(NConv1, 5, activation='relu', input_shape=(240, 6)))
model6.add(Dropout(0.2))
model6.add(Conv1D(NConv2, 5, activation='relu'))
model6.add(Dropout(0.2))
model6.add(JANET(NJanet1, activation='relu', return_sequences=True))
model6.add(Dropout(0.2))
model6.add(JANET(NJanet2, activation='relu'))
model6.add(Dropout(0.2))
model6.add(Dense(NDense1, activation='relu'))
model6.add(Dropout(0.2))
model6.add(Dense(NDense2, activation='relu'))
model6.add(Dropout(0.2))
model6.add(Dense(20, activation='softmax'))
model6.compile(loss='categorical_crossentropy',
optimizer=nadam,
metrics=['accuracy'])
model6.fit(xTrain,
yTrain,
epochs=epochs,
validation_data=(xVal, yVal),
acc_list = []
prec_list = []
recall_list = []
f1_list = []
for i in range(repetition):
early_stopping = EarlyStopping(monitor='val_acc', patience=patience)
checkpoint = ModelCheckpoint('6dmg_try.h5',
monitor='val_acc',
verbose=0,
save_weights_only=True)
nadam = Nadam(lr=1e-4, beta_1=0.9, beta_2=0.9)
model7 = Sequential()
model7.add(
JANET(NJanet1,
activation='relu',
return_sequences=True,
input_shape=(None, 6)))
model7.add(Dropout(0.2))
model7.add(JANET(NJanet2, activation='relu'))
model7.add(Dropout(0.2))
model7.add(Dense(NDense1, activation='relu'))
model7.add(Dropout(0.2))
model7.add(Dense(NDense2, activation='relu'))
model7.add(Dropout(0.2))
model7.add(Dense(20, activation='softmax'))
model7.compile(loss='categorical_crossentropy',
optimizer=nadam,
metrics=['accuracy'])
model7.fit(xTrain,
yTrain,
epochs=epochs,
add_indices[i, [first_half, second_half]] = 1.
# Zip the values and indices in a third dimension:
# inputs has the shape (batch_size, time_steps, 2)
inputs = np.dstack((add_values, add_indices))
targets = np.sum(np.multiply(add_values, add_indices), axis=1)
# center at zero mean
inputs -= np.mean(inputs, axis=0, keepdims=True)
yield inputs, targets
print('Build model...')
model = Sequential()
model.add(JANET(NUM_UNITS, input_shape=(TIME_STEPS, 2)))
model.add(Dense(1, activation='linear'))
# try using different optimizers and different optimizer configs
optimizer = Adam(LEARNING_RATE, amsgrad=True)
model.compile(loss='mse', optimizer='adam')
model.fit_generator(batch_generator(),
steps_per_epoch=STEPS_PER_EPOCH,
epochs=NUM_EPOCHS,
verbose=1,
callbacks=[
ModelCheckpoint('weights/janet_addition_%d.h5' %
(TIME_STEPS),
monitor='loss',
save_best_only=True,
print('Loading data...')
(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=max_features)
print(len(x_train), 'train sequences')
print(len(x_test), 'test sequences')
print('Pad sequences (samples x time)')
x_train = sequence.pad_sequences(x_train, maxlen=maxlen)
x_test = sequence.pad_sequences(x_test, maxlen=maxlen)
print('x_train shape:', x_train.shape)
print('x_test shape:', x_test.shape)
print('Build model...')
model = Sequential()
model.add(Embedding(max_features, 128, input_shape=(maxlen, )))
model.add(JANET(128, dropout=0.0, recurrent_dropout=0.0))
model.add(Dense(1, activation='sigmoid'))
# try using different optimizers and different optimizer configs
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.summary()
print('Train...')
model.fit(x_train,
y_train,
batch_size=batch_size,
epochs=15,
validation_data=(x_test, y_test),
x_train = (x_train - x_train.mean(axis=0, keepdims=True)) / (
x_train.std(axis=0, keepdims=True) + 1e-8)
x_test = (x_test - x_test.mean(axis=0, keepdims=True)) / (
x_test.std(axis=0, keepdims=True) + 1e-8)
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
print('Evaluate IRNN...')
model = Sequential()
model.add(JANET(hidden_units, input_shape=x_train.shape[1:]))
model.add(Dense(num_classes, activation='softmax'))
model.summary()
rmsprop = Adam(lr=learning_rate, amsgrad=True)
model.compile(loss='categorical_crossentropy',
optimizer=rmsprop,
metrics=['accuracy'])
# model.fit(x_train, y_train,
# batch_size=batch_size,
# epochs=epochs,
# verbose=1,
# validation_data=(x_test, y_test),