counts = counts.to(device)
ids = ids.to(device)
batch_size = cw_idxs.size(0)
# Forward
output_dict = model(cw_idxs, cc_idxs, qw_idxs, qc_idxs, ids,
start_idxs, end_idxs, counts)
loss = output_dict['loss']
nll_meter.update(loss.item(), batch_size)
# Log info
progress_bar.update(batch_size)
progress_bar.set_postfix(NLL=nll_meter.avg)
pred_dict.update(output_dict["predictions"])
model.set_eval_data(None)
model.train()
# Get F1 and EM scores
eval_dict = eval_dicts(gold_dict, pred_dict)
results_list = [('Loss', nll_meter.avg), ('F1', eval_dict['F1']),
('EM', eval_dict['EM'])]
eval_dict = OrderedDict(results_list)
return eval_dict, pred_dict
if __name__ == '__main__':
main(get_train_args())
question_char_idxs = merge_2d(question_char_idxs)
# number_indices = merge_1d(number_indices, pad_value = -1)
start_indices = merge_1d(start_indices, pad_value=-1)
end_indices = merge_1d(end_indices, pad_value=-1)
counts = merge_0d(counts) # TODO check
# add_sub_expressions = merge_2d(add_sub_expressions, pad_value = -1)
ids = merge_0d(ids)
return (context_idxs, context_char_idxs, question_idxs, question_char_idxs,
start_indices, end_indices, counts, ids)
if __name__ == "__main__":
args = get_train_args()
torch.manual_seed(224)
print("Building datasets...", end=" ")
train_dataset = DROP(args.train_record_file)
train_loader = data.DataLoader(train_dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.num_workers,
collate_fn=collate_fn)
dev_dataset = DROP(args.dev_record_file)
dev_loader = data.DataLoader(dev_dataset,
batch_size=args.batch_size,
shuffle=False,
num_workers=args.num_workers,
collate_fn=collate_fn)
print("Done!")