utils.py

import os
import torch
import random
import logging
import numpy as np
from models import JointTinyBert2, JointBERT
from seqeval.metrics import precision_score, recall_score, f1_score


def get_args(model_dir):
    return torch.load('{}/training_args.bin'.format(model_dir))


def load_model(model_dir, args, device):
    # Check whether model exists
    if not os.path.exists(model_dir):
        raise Exception("Model doesn't exists! Train first!")

    try:
        model = JointTinyBert2.from_pretrained(model_dir,
                                               args=args,
                                               intent_label_lst=get_intent_labels(args),
                                               slot_label_lst=get_slot_labels(args))
        model.to(device)
        model.eval()
        print("***** Model Loaded *****")
    except:
        raise Exception("Some model files might be missing...")

    return model


def init_logger():
    logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s -   %(message)s',
                        datefmt='%m/%d/%Y %H:%M:%S',
                        level=logging.INFO)


def get_intent_labels(args):
    return [label.strip() for label in open(os.path.join(
        args.data_dir, args.task, args.intent_label_file), 'r', encoding='utf-8')]


def get_slot_labels(args):
    return [label.strip() for label in open(os.path.join(
        args.data_dir, args.task, args.slot_label_file), 'r', encoding='utf-8')]


def set_seed(args):
    random.seed(args.seed)
    np.random.seed(args.seed)
    os.environ['PYTHONHASHSEED'] = str(args.seed)
    torch.manual_seed(args.seed)
    if not args.no_cuda and torch.cuda.is_available():
        torch.cuda.manual_seed_all(args.seed)


def compute_metrics(intent_preds, intent_labels, slot_preds, slot_labels):
    assert len(intent_preds) == len(intent_labels) == len(slot_preds) == len(slot_labels)
    results = {}
    intent_result = get_intent_acc(intent_preds, intent_labels)
    slot_result = get_slot_metrics(slot_preds, slot_labels)
    sementic_result = get_sentence_frame_acc(intent_preds, intent_labels, slot_preds, slot_labels)

    results.update(intent_result)
    results.update(slot_result)
    results.update(sementic_result)

    return results


def get_slot_metrics(preds, labels):
    assert len(preds) == len(labels)
    return {
        "slot_precision": precision_score(labels, preds),
        "slot_recall": recall_score(labels, preds),
        "slot_f1": f1_score(labels, preds)
    }


def get_intent_acc(preds, labels):
    acc = (preds == labels).mean()
    return {
        "intent_acc": acc
    }


def get_sentence_frame_acc(intent_preds, intent_labels, slot_preds, slot_labels):
    """For the cases that intent and all the slots are correct (in one sentence)"""
    # Get the intent comparison result
    intent_result = (intent_preds == intent_labels)

    # Get the slot comparision result
    slot_result = []
    for preds, labels in zip(slot_preds, slot_labels):
        assert len(preds) == len(labels)
        one_sent_result = True
        for p, l in zip(preds, labels):
            if p != l:
                one_sent_result = False
                break
        slot_result.append(one_sent_result)
    slot_result = np.array(slot_result)

    sementic_acc = np.multiply(intent_result, slot_result).mean()
    return {
        "sementic_frame_acc": sementic_acc
    }