elif model == 'lstmbm':
Model = LSTM_multilayers
weight_file = 'LSTM_multi5.model'
elif model == 'grnb1':
Model = GRN_bitmap
weight_file = 'GRN_bitmap5.model'
elif model == 'grnbm':
Model = GRN_multilayer
weight_file = 'GRN_multi5.model'
model_network = Model.get_model()
model_network.load_weights(weight_file)
iterator = prepare_data(validate_x_file, validate_y_file)
results = []
for validate_x, validate_y in iterator:
predict_y = model_network.predict_classes(validate_x)
shape_y = predict_y.shape
predict_y = predict_y.reshape((shape_y[0] * shape_y[1], shape_y[2]))
predict_y = (predict_y == [0])[:, 0]
validate_y = validate_y.reshape((shape_y[0] * shape_y[1], shape_y[2]))
validate_y = (validate_y == [0])[:, 0]
a = precision_recall_fscore_support(validate_y, predict_y, average='binary')
print(a)
results.append(a[:3])
results = np.array(results)
print('final result:')
model.add_input(name='input', input_shape=(200, 10*5))
model.add_node(LSTM(128, return_sequences=True, input_shape=(200, 10*5)),
name='forward', input='input')
model.add_node(LSTM(128, return_sequences=True, input_shape=(200, 10*5), go_backwards=True),
name='backward', input='input')
model.add_node(Dropout(0.2), name='dropout', inputs=['forward', 'backward'])
model.add_node(TimeDistributedDense(2, activation='sigmoid'), name='sigmoid', input='dropout')
model.add_output(name='output', input='sigmoid')
print('compile')
model.compile('adam', {'output': 'binary_crossentropy'})
return model
if __name__ == '__main__':
train_x_file = '../' + sys.argv[1]
train_y_file = '../' + sys.argv[2]
modelfile = sys.argv[3]
# test_x_file = '../' + sys.argv[3]
# test_y_file = '../' + sys.argv[4]
print('loaded')
model = get_model()
print("Train...")
for epoch in range(2):
iterator = prepare_data(train_x_file, train_y_file)
print('epoch:', epoch)
for x, y in iterator:
model.fit({'input': x, 'output': y}, nb_epoch=1)
model.save_weights(modelfile)
