print('Best parameters:')
print(grid.best_params_)
results = pd.DataFrame(grid.cv_results_)
results.sort_values(by='rank_test_score', inplace=True)
print(results.to_string())
# Build top 3 models
idx = 0
top_models = []
for index, row in results.iterrows():
top_model = False
if (single_branch):
top_model = cnn_batchnorm_lstm.build_model(
row['param_layers'], row['param_cnn_layers'],
row['param_kernel_size'])
else:
top_model = bidir_multiple_branch_lstm.build_model(
row['param_layers'], row['param_ksize1'],
row['param_step_size'], row['param_lstm_units'],
row['param_single_lstm_layer'])
top_models.append(top_model)
top_models[-1].batch_size = row['param_batch_size']
idx = idx + 1
if idx > 2:
break
early_stopping = EarlyStopping(patience=20)
fit_models = []
for i in reversed(range(len(top_models))):
if __name__=='__main__':
global_start_time = time.time()
#epochs = 1
epochs = 100
seq_len = 40
print('> Loading data... ')
X_train, y_train, X_test, y_test = dataload.load_data('daily_spx.csv', seq_len, True)
print('> Data Loaded. Compiling...')
#lstm_model = lstm.build_model([1, seq_len, 100, 1])
model_layers = [1, 2, 3, 4]
cnn_models = [
cnn_batchnorm_lstm.build_model([1, seq_len, 100, 1], x) for x in model_layers
]
[model.fit(
X_train,
y_train,
batch_size=512,
nb_epoch=epochs,
validation_split=0.05)
for model in cnn_models
]
#lstm_predictions = dataload.predict_sequences_multiple(lstm_model, X_test, seq_len, seq_len)
cnn_predictions = [dataload.predict_sequences_multiple(model, X_test, seq_len, seq_len)
for model in cnn_models]
#predictions = dataload.predict_sequence_full(model, X_test, seq_len)
#predictions = dataload.predict_point_by_point(model, X_test)