# "rate_of_learning": 0.26343183932470754,
# "max_epoch_size": 3,
# "gaussian_noise_stdev": 0.0007517656514955944,
# "l2_regularization": 0.00022259525510874703,
# "max_num_epochs": 14,
# "random_normal_initializer_stdev": 0.0005827304210740794
# }
# cif
# optimized_configuration = {'num_hidden_layers': 5.0, 'cell_dimension': 28.471127262736434,
# 'minibatch_size': 10.135034205224617, 'max_epoch_size': 9.1502825822926326,
# 'max_num_epochs': 20.962475980675006, 'l2_regularization': 0.0006369387641617046,
# 'gaussian_noise_stdev': 0.00057001364478555087,
# 'random_normal_initializer_stdev': 0.00025797511482927632,
# 'rate_of_learning': 0.20172634121590136}
# persist the optimized configuration to a file
persist_results(
optimized_configuration,
optimized_config_directory + '/' + model_identifier + '.txt')
#optimized_configuration = read_optimal_hyperparameter_values(optimized_config_directory + '/' + model_identifier + '.txt')
# test the model
#for i in range(1, 11):
# args.seed = i
testing(args, optimized_configuration)
end = time.time()
print(end - start)
count = 0
for lag in lags:
current_lag = lag
if (require_validation):
range_of_series = [i for i in range(0, num_of_series)]
# Randomly choose 1/7 series and tune hyperparameters for that
chosen_indices = random.sample(range_of_series,
int(len(range_of_series) * (1 / 7)))
optimized_configuration = smac()
# persist the optimized configuration to a file
persist_results(
optimized_configuration,
base_dir + optimized_config_directory + "fnn_" + dataset_name +
'_lag_' + str(lag) + '.txt')
validation_error = train_model(optimized_configuration)
print(optimized_configuration)
print(validation_error)
else:
# Calculate forecasts FFNN forecasts for the original dataset using a chosen clustering approach
subprocess.call([
"Rscript", "--vanilla",
base_dir + "feed_forward_nn/ffnn_cluster_tester.R",
str(num_of_hidden_nodes[count]),
str(decay[count]), original_file, results_file, dataset_name,
str(horizon),
str(lag),
str(address_near_zero_insability),
str(integer_conversion),