[p for n, p in param_optimizer if any(nd in n for nd in no_decay)],
"weight_decay":
0.0,
},
]
# Optimizer and Scheduler
optimizer = torch.optim.AdamW(optimizer_parameters, lr=3e-5)
num_training_steps = int(
len(train_dataset) / config.TRAIN_BATCH_SIZE * config.EPOCHS)
scheduler = get_linear_schedule_with_warmup(
optimizer, num_warmup_steps=0, num_training_steps=num_training_steps)
# Training loop
best_micro = 0
for epoch in range(config.EPOCHS):
engine.train(epoch, model, training_loader, device, optimizer,
scheduler)
outputs, targets = engine.validation(epoch, model, testing_loader,
device)
outputs = np.array(outputs) >= 0.5
accuracy = metrics.accuracy_score(targets, outputs)
f1_score_micro = metrics.f1_score(targets, outputs, average='micro')
f1_score_macro = metrics.f1_score(targets, outputs, average='macro')
print(f"Accuracy Score = {accuracy}")
print(f"F1 Score (Micro) = {f1_score_micro}")
print(f"F1 Score (Macro) = {f1_score_macro}")
if f1_score_micro > best_micro:
torch.save(model.state_dict(), config.MODEL_PATH)
best_micro = f1_score_micro
train_loader,
model,
train_dir,
ord_loss,
optimizer,
epoch,
logger,
PRINT_FREQ,
BETA=BETA,
GAMMA=GAMMA,
ORD_NUM=80.0)
validation(device,
val_loader,
model,
ord_loss,
val_dir,
epoch,
logger,
PRINT_FREQ,
BETA=BETA,
GAMMA=GAMMA,
ORD_NUM=80.0)
# save model and checkpoint per epoch
checkpoint_filename = os.path.join(
output_dir, 'checkpoint-{}.pth.tar'.format(str(epoch)))
torch.save(model, checkpoint_filename)
epochbar.update(1)