def run(self, iterator, criterion, optimizer, verbose=True):
r"""Train one time the model on iterator data.
Args:
iterator (Iterator): iterator containing batch samples of data.
criterion (Loss): loss function to measure scores.
optimizer (Optimizer): optimizer used during training to update weights.
verbose (bool): if ``True`` display a progress bar.
Returns:
dict: the performance and metrics of the training session.
"""
# Initialize the variables
start_time = time.time()
epoch_loss = 0
epoch_acc = 0
class_labels = list(range(self.output_dim))
class_labels.pop(self.pad_idx_label)
confusion_matrix = ConfusionMatrix(labels=class_labels)
# Train mode
self.train()
for (idx, batch) in enumerate(iterator):
optimizer.zero_grad()
# One forward step
text, length = batch.text
y_hat = self.forward(text, length)
y_hat = y_hat.view(-1, y_hat.shape[-1])
label = batch.label.view(-1)
# Get the predicted classes
y_tilde = y_hat.argmax(dim=1, keepdim=True)
# Compute the loss and update the weights
loss = criterion(y_hat, label)
loss.backward()
optimizer.step()
epoch_loss += loss.item()
# Default accuracy
acc = self.get_accuracy(y_tilde, label)
epoch_acc += acc.item()
# Optional: display a progress bar
if verbose:
progress_bar(idx, len(iterator) - 1, prefix="Training:\t", start_time=start_time)
# Update the confusion matrix
confusion_matrix.update(label.long().numpy(), y_tilde.long().numpy())
# Store the loss, accuracy and metrics in a dictionary
results_train = {"loss": epoch_loss / len(iterator),
"accuracy": epoch_acc / len(iterator),
**confusion_matrix.to_dict()
}
return results_train
def download_from_url(url, save_path):
"""Download a file from an URL.
Args:
url (str): path to the URL.
save_path (str): path to the saving directory.
Returns:
None
"""
response = requests.get(url, stream=True)
total = response.headers.get('content-length')
with open(save_path, 'wb') as f:
if total is None:
f.write(response.content)
else:
downloaded = 0
total = int(total)
for data in response.iter_content(chunk_size=max(int(total / 1000), 1024 * 1024)):
downloaded += len(data)
f.write(data)
progress_bar(downloaded, total, prefix="Downloading...")