def get_accuracies(file_to_print):
logger = logging.getLogger(CONFIG['experiment_name'])
configure_logger_by_default(logger)
logger.info("START " + CONFIG['experiment_name'])
def print_info(info):
logger.info(info)
print(info)
file_to_print.write(info + "\n")
CONFIG['pso_options']['print_info'] = print_info
train_index, test_index = next(CONFIG['cv'].split(INPUT_FEATURES,
INPUT_LABELS))
train_features, train_labels = INPUT_FEATURES[train_index], INPUT_LABELS[
train_index]
test_features, test_labels = INPUT_FEATURES[test_index], INPUT_LABELS[
test_index]
scaler = preprocessing.StandardScaler().fit(train_features)
train_features = scaler.transform(train_features)
test_features = scaler.transform(test_features)
train_features = torch.from_numpy(train_features)
train_labels = torch.from_numpy(train_labels)
test_features = torch.from_numpy(test_features)
test_labels = torch.from_numpy(test_labels)
train_accuracies = []
test_accuracies = []
hidden_dims = []
for hidden_layer_percentage in CONFIG['hidden_layer_percentage']:
hidden_dim = int(CONFIG['max_hidden_layer_size'] *
hidden_layer_percentage)
model = Model(hidden_dim=hidden_dim)
test_acc, train_acc = train_bp(model, train_features, train_labels,
test_features, test_labels, CONFIG)
train_accuracies.append(train_acc)
test_accuracies.append(test_acc)
hidden_dims.append(hidden_dim)
print_info("======RESULTING ACCURACIES======")
print_info("TRAIN: " + str(train_accuracies))
print_info("TEST: " + str(test_accuracies))
print_info("HIDDEN DIMS: " + str(hidden_dims))
logger.info("END " + CONFIG['experiment_name'])
return train_accuracies, test_accuracies, hidden_dims
def get_accuracies(file_to_print):
logger = logging.getLogger(CONFIG['experiment_name'])
configure_logger_by_default(logger)
logger.info("START " + CONFIG['experiment_name'])
def print_info(info):
logger.info(info)
print(info)
file_to_print.write(info + "\n")
CONFIG['pso_options']['print_info'] = print_info
full_train_index, test_index = next(CONFIG['cv'].split(
INPUT_FEATURES, INPUT_LABELS))
np.random.shuffle(full_train_index)
train_accuracies = []
test_accuracies = []
for train_percentage in CONFIG['train_splits']:
train_size = int(train_percentage * full_train_index.shape[0])
train_index = full_train_index[0:train_size]
train_features, train_labels = INPUT_FEATURES[
train_index], INPUT_LABELS[train_index]
test_features, test_labels = INPUT_FEATURES[test_index], INPUT_LABELS[
test_index]
scaler = preprocessing.StandardScaler().fit(train_features)
train_features = scaler.transform(train_features)
test_features = scaler.transform(test_features)
train_features = torch.from_numpy(train_features)
train_labels = torch.from_numpy(train_labels)
test_features = torch.from_numpy(test_features)
test_labels = torch.from_numpy(test_labels)
model = Model()
test_acc, train_acc = train_bp(model, train_features, train_labels,
test_features, test_labels, CONFIG)
train_accuracies.append(train_acc)
test_accuracies.append(test_acc)
print_info("======RESULTING ACCURACIES======")
print_info("TRAIN: " + str(train_accuracies))
print_info("TEST: " + str(test_accuracies))
logger.info("END " + CONFIG['experiment_name'])
return train_accuracies, test_accuracies
def fit_model(file_to_print):
logger = logging.getLogger('one_split_fit')
configure_logger_by_default(logger)
logger.info("START fit_model")
def print_info(info):
logger.info(info)
print(info)
file_to_print.write(info + "\n")
CONFIG['pso_options']['print_info'] = print_info
train_index, test_index = next(CONFIG['cv'].split(INPUT_FEATURES,
INPUT_LABELS))
train_features, train_labels = INPUT_FEATURES[train_index], INPUT_LABELS[
train_index]
test_features, test_labels = INPUT_FEATURES[test_index], INPUT_LABELS[
test_index]
scaler = preprocessing.StandardScaler().fit(train_features)
train_features = scaler.transform(train_features)
test_features = scaler.transform(test_features)
train_features = torch.from_numpy(train_features)
train_labels = torch.from_numpy(train_labels)
test_features = torch.from_numpy(test_features)
test_labels = torch.from_numpy(test_labels)
model = Model()
for trainer_type in CONFIG['trainers_to_use']:
if trainer_type == 'bp':
test_acc, train_acc, test_loss, train_loss = \
train_bp(model, train_features, train_labels, test_features, test_labels, CONFIG)
print_info("Best Test Accuracy: " + str(test_acc) +
", Train Accuracy: " + str(train_acc) +
", Test Loss: " + str(test_loss) + ", Train Loss: " +
str(train_loss))
elif trainer_type == 'pso':
train_pso(model, train_features, train_labels, test_features,
test_labels, CONFIG)
logger.info("END fit_model")
def run_cross_validation_psobp(file_to_print) -> torch.Tensor:
logger = logging.getLogger('10_fold_cv')
configure_logger_by_default(logger)
logger.info("START run_cross_validation")
def print_info(info):
logger.info(info)
print(info)
file_to_print.write(info + "\n")
accuracies = torch.zeros((CONFIG['n_splits'], CONFIG['n_repeats']))
fold_number = -1
repeat_number = -1
CONFIG['pso_options']['print_info'] = print_info
for train_index, test_index in CONFIG['cv'].split(INPUT_FEATURES,
INPUT_LABELS):
fold_number += 1
if fold_number % CONFIG['n_splits'] == 0:
fold_number = 0
repeat_number += 1
def print_info(info):
logger.info(info)
print(info)
file_to_print.write(info + "\n")
print_info("New " + str(fold_number) + " fold. repeat = " +
str(repeat_number))
train_features, train_labels = INPUT_FEATURES[
train_index], INPUT_LABELS[train_index]
test_features, test_labels = INPUT_FEATURES[test_index], INPUT_LABELS[
test_index]
scaler = preprocessing.StandardScaler().fit(train_features)
train_features = scaler.transform(train_features)
test_features = scaler.transform(test_features)
train_features = torch.from_numpy(train_features)
train_labels = torch.from_numpy(train_labels)
test_features = torch.from_numpy(test_features)
test_labels = torch.from_numpy(test_labels)
model = Model().to(CONFIG['device'])
best_accuracy_bp, best_accuracy_pso = None, None
for trainer_type in CONFIG['trainers_to_use']:
if trainer_type == 'bp':
best_accuracy_bp, _ = train_bp(model, train_features,
train_labels, test_features,
test_labels, CONFIG)
elif trainer_type == 'pso':
best_accuracy_pso = train_pso(model, train_features,
train_labels, test_features,
test_labels, CONFIG)
print_info('Best accuracy pso: ' + str(best_accuracy_pso) + ', bp: ' +
str(best_accuracy_bp))
best_accuracy = max(best_accuracy_bp, best_accuracy_pso)
accuracies[fold_number][repeat_number] = best_accuracy
print_info('Best final accuracy: ' + str(best_accuracy))
print_info('END OF EVALUATION OF ' + str(fold_number) +
' FOLD, REPEAT ' + str(repeat_number))
print_info(
'========================================================================='
)
logger.info("END run_cross_validation")
return accuracies