def qanlu_rec_to_conll()

in calculate_pr.py [0:0]

• 42 lines of code
• 12 McCabe index (conditional complexity)

def qanlu_rec_to_conll(rec, predictions_):
    context = rec['context']
    slots = []
    preds = []

    for qa in rec['qas']:
        if not qa['id'].startswith("intent_") and qa['answers']:
            slots.append(
                {'slot': qa['slot'], 'ans': qa['answers'][0]['text'], 'start': qa['answers'][0]['answer_start']})

        pred_ans = predictions_[qa['id']]

        if not qa['id'].startswith("intent_") and pred_ans:
            preds.append({'slot': qa['slot'], 'ans': pred_ans, 'start': context.find(pred_ans)})

    slots = sorted(slots, key=lambda i: i['start'], reverse=False)
    preds = sorted(preds, key=lambda i: i['start'], reverse=False)

    prev = 0
    labels = []
    for slot in slots:
        curr = slot['start']
        prev_sent = context[prev:curr]
        labels.extend(['O'] * len(prev_sent.split()))
        curr_slot_split = slot['ans'].split()
        labels.append('B-' + slot['slot'])
        if len(curr_slot_split) > 1:
            for _ in curr_slot_split[1:]:
                labels.append('I-' + slot['slot'])
        prev = slot['start'] + len(slot['ans']) + 1

    labels.extend(['O'] * len(context[prev:].split()))
    y_true = labels

    prev = 0
    labels = []
    for slot in preds:
        curr = slot['start']
        prev_sent = context[prev:curr]
        labels.extend(['O'] * len(prev_sent.split()))
        curr_slot_split = slot['ans'].split()
        labels.append('B-' + slot['slot'])

        if len(curr_slot_split) > 1:
            for _ in curr_slot_split[1:]:
                labels.append('I-' + slot['slot'])

        prev = slot['start'] + len(slot['ans']) + 1

    labels.extend(['O'] * len(context[prev:].split()))
    y_pred = labels

    return y_true, y_pred, context.split()