def run_baseline_batch()

in scripts/run_model.py [0:0]

• 26 lines of code
• 3 McCabe index (conditional complexity)

def run_baseline_batch(args, model, loader, dtype):
  model.type(dtype)
  model.eval()

  all_scores, all_probs = [], []
  num_correct, num_samples = 0, 0
  for batch in loader:
    questions, images, feats, answers, programs, program_lists = batch

    questions_var = Variable(questions.type(dtype).long(), volatile=True)
    feats_var = Variable(feats.type(dtype), volatile=True)
    scores = model(questions_var, feats_var)
    probs = F.softmax(scores)

    _, preds = scores.data.cpu().max(1)
    all_scores.append(scores.data.cpu().clone())
    all_probs.append(probs.data.cpu().clone())

    num_correct += (preds == answers).sum()
    num_samples += preds.size(0)
    print('Ran %d samples' % num_samples)

  acc = float(num_correct) / num_samples
  print('Got %d / %d = %.2f correct' % (num_correct, num_samples, 100 * acc))

  all_scores = torch.cat(all_scores, 0)
  all_probs = torch.cat(all_probs, 0)
  if args.output_h5 is not None:
    print('Writing output to %s' % args.output_h5)
    with h5py.File(args.output_h5, 'w') as fout:
      fout.create_dataset('scores', data=all_scores.numpy())
      fout.create_dataset('probs', data=all_probs.numpy())