Notes

import numpy as np
# function to get validation accuracy
def get_validation_performance(val_set):
# Put the model in evaluation mode
model.eval()

# Tracking variables
total_correct = 0
total_eval_loss = 0


num_batches = int(len(val_set)/batch_size)
if int(len(val_set)/batch_size)!=len(val_set)/batch_size:
num_batches+=1

for i in range(num_batches):
end_index = min(batch_size * (i+1), len(val_set))

batch = val_set[i*batch_size:end_index]

input_id_tensors = torch.stack([data[0] for data in batch])
input_mask_tensors = torch.stack([data[1] for data in batch])
label_tensors = torch.stack([data[2] for data in batch])

# Move tensors to the GPU
b_input_ids = input_id_tensors.to(device)
b_input_mask = input_mask_tensors.to(device)
b_labels = label_tensors.to(device)

# Tell pytorch not to bother with constructing the compute graph during
# the forward pass, since this is only needed for backprop (training).
with torch.no_grad():

# Forward pass, calculate logit predictions.
outputs = model(b_input_ids,token_type_ids=None,attention_mask=b_input_mask,labels=b_labels)
loss = outputs.loss
logits = outputs.logits

# Accumulate the validation loss.
total_eval_loss += loss.item()

# Move logits and labels to CPU
logits = logits.detach().cpu().numpy()
label_ids = b_labels.to('cpu').numpy()


# Calculate the number of correctly labeled examples in batch
pred_flat = np.argmax(logits, axis=1).flatten()
labels_flat = label_ids.flatten()
num_correct = np.sum(pred_flat == labels_flat)
total_correct += num_correct

# Report the final accuracy for this validation run.
avg_val_accuracy = total_correct / len(val_set)
return avg_val_accuracy