test_name = test_name + "_data-" + dataset_name + "_nFold-" + str(n_folds) + "_lr-" + \
str(lr) +"_drop_prob-"+str(drop_prob)+"_weight-decay-"+ str(weight_decay)+ \
"_batchSize-" + str(batch_size) + "_nHidden-" + str(n_units) + \
"_output-" + str(output) + "_maxK-" + str(max_k)
training_log_dir = os.path.join("./test_log/", test_name)
if not os.path.exists(training_log_dir):
os.makedirs(training_log_dir)
printParOnFile(test_name=test_name, log_dir=training_log_dir,
par_list={"dataset_name": dataset_name,
"n_fold": n_folds,
"learning_rate": lr,
"drop_prob": drop_prob,
"weight_decay": weight_decay,
"batch_size": batch_size,
"n_hidden": n_units,
"test_epoch": test_epoch,
"output": output,
"max_k": max_k})
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
criterion = torch.nn.NLLLoss()
dataset_cv_splits = getcross_validation_split(dataset_path, dataset_name, n_folds, batch_size)
for split_id, split in enumerate(dataset_cv_splits):
loader_train = split[0]
loader_test = split[1]
loader_valid = split[2]
"_som_grid-" + str(som_grids_dim[0]) + "_" + str(som_grids_dim[1]) + \
"_som_lr-" + str(som_lr)
training_log_dir = os.path.join("./test_log/", test_name)
if not os.path.exists(training_log_dir):
os.makedirs(training_log_dir)
printParOnFile(test_name=test_name,
log_dir=training_log_dir,
par_list={
"dataset_name": dataset_name,
"n_fold": n_folds,
"learning_rate_conv": lr_conv,
"learning_rate_som": som_lr,
"learning_rate_read_out": lr_readout,
"learning_rate_fine_tuning": lr_fine_tuning,
"drop_prob": drop_prob,
"weight_decay": weight_decay,
"batch_size": batch_size,
"n_hidden": n_units,
"som_grid_dims": som_grids_dim,
"som_lr": som_lr,
"test_epoch": test_epoch
})
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
criterion = torch.nn.NLLLoss()
dataset_cv_splits = getcross_validation_split(dataset_path, dataset_name,
n_folds, batch_size)
