INNER CODE UNIT · Python

padding

IBM/transition-amr-parser · src/ibm_neural_aligner/main.py:602

            padding = [PADDING_IDX] * (max_length - len(v))
            bos = [PADDING_IDX]
            val.append(bos + v + padding)

        batch_map[k] = torch.tensor(val, dtype=dtypes[k], device=device)

    batch_map['text_original_tokens'] = [x['text_original_tokens'] for x in items]
    batch_map['items'] = items
    batch_map['device'] = device
    batch_map['add_edges'] = args.add_edges

    if args.mask > 0 and train:
        batch_mask = []
        batch_mask_gcn = []
        for x in items:
            tokens = [MaskInfo.unchanged, MaskInfo.masked]
            probs = [1 - args.mask, args.mask]
            size = len(x['amr_nodes'])

View source record →