def main(dl: DataLoader, model: RNN):
prev_best = 0
patience = 0
decay = 0
lr = config.lr
#optimizer = torch.optim.SGD(model.parameters(), lr=lr)
optimizer = torch.optim.Adam(model.parameters(), lr=lr)
criterion = nn.BCEWithLogitsLoss()
for epoch in tqdm(range(config.max_epochs)):
start_time = time.time()
train_loss, train_acc = train(dl.train_examples, model, optimizer,
criterion)
dev_loss, dev_acc = evaluate(dl.dev_examples, model, criterion)
end_time = time.time()
epoch_mins, epoch_secs = epoch_time(start_time, end_time)
print(
f'Epoch: {epoch + 1:02} | Epoch Time: {epoch_mins}m {epoch_secs}s')
print(f'\tTrain Loss: {train_loss:.3f} | Train Acc: {train_acc:.2f}')
print(f'\t Dev Loss: {dev_loss:.3f} | Dev Acc: {dev_acc:.2f}')
if dev_acc <= prev_best:
patience += 1
if patience == 3:
lr *= 0.5
optimizer = torch.optim.SGD(model.parameters(), lr=lr)
tqdm.write(
'Dev accuracy did not increase in 3 epochs, halfing the learning rate'
)
patience = 0
decay += 1
else:
prev_best = dev_acc
print('Save the best model')
model.save()
if decay >= 3:
print('Evaluating model on test set')
model.load()
print('Load the best model')
test_loss, test_acc = evaluate(dl.test_examples, model, criterion)
print(
f'Test Loss: {test_loss:.3f} | Test Acc: {test_acc * 100:.2f}%'
)
break
if __name__ == '__main__':
# people can find their devices using tf.config.list_physical_devices("GPU") # "CPU"
with tf.device("/GPU:0"):
# Model Training
model.fit(
imdb.X_train,
imdb.Y_train,
batch_size=512,
epochs=10,
validation_split=0.2,
callbacks=[EarlyStopping(patience=2, verbose=1)],
)
# Run model on test set
accr = model.evaluate(imdb.X_test, imdb.Y_test)
print('Test set\n Loss: {:0.4f}\n Accuracy: {:0.2f}'.format(
accr[0], accr[1] * 100))
# save weights as HDF5
model.save("../model/tf/weights.h5")
print("Saved model to disk")
# save model as JSON
model_json = model.to_json()
with open("../model/tf/model.json", "w") as file:
file.write(model_json)
# save tokenizer as JSON
tokenizer_json = imdb.tokenizer.to_json()
with open("../model/tf/tokenizer.json", 'w', encoding='utf-8') as file:
file.write(json.dumps(tokenizer_json, ensure_ascii=True))
