def evaluate()

in src/baselines/dnn.py [0:0]

• 73 lines of code
• 13 McCabe index (conditional complexity)

def evaluate(args, model, dataset, prefix, log_fp=None):
    '''
    evaluate
    :param args:
    :param model:
    :param dataset:
    :param prefix:
    :param log_fp:
    :return:
    '''
    output_dir = os.path.join(args.output_dir, prefix)
    print("Evaluating dir path: ", output_dir)
    if not os.path.exists(output_dir) and args.local_rank in [-1, 0]:
        os.makedirs(output_dir)
    result = {}
    args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu)

    # Note that DistributedSampler samples randomly
    eval_sampler = SequentialSampler(dataset)
    eval_dataset_total_num = len(dataset)
    eval_dataloader = DataLoader(dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
    eval_batch_total_num = len(eval_dataloader)
    print("Dev dataset len: %d, batch num: %d" % (eval_dataset_total_num, eval_batch_total_num))

    # multi GPU
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # evaluate
    logger.info("***** Running evaluation {} *****".format(prefix))
    logger.info("Num examples = %d", eval_dataset_total_num)
    logger.info("Batch size = %d", args.eval_batch_size)
    if log_fp:
        log_fp.write("***** Running evaluation {} *****\n".format(prefix))
        log_fp.write("Dev Dataset Num examples = %d\n" % eval_dataset_total_num)
        log_fp.write("Dev Dataset Instantaneous batch size per GPU = %d\n" % args.per_gpu_eval_batch_size)
        log_fp.write("Dev Dataset batch number = %d\n" % eval_batch_total_num)
        log_fp.write("#" * 50 + "\n")
    eval_loss = 0.0
    nb_eval_steps = 0
    # predicted prob
    pred_scores = None
    # ground truth
    out_label_ids = None

    for batch in tqdm(eval_dataloader, total=eval_batch_total_num, desc="Evaluating"):
        # evaluate
        model.eval()
        batch = tuple(t.to(args.device) for t in batch)
        with torch.no_grad():
            inputs = {
                "inputs": batch[0],
                "labels": batch[-1]
            }
            outputs = model(**inputs)
            tmp_eval_loss, logits, output = outputs[:3]

            eval_loss += tmp_eval_loss.mean().item()
        nb_eval_steps += 1
        if pred_scores is None:
            pred_scores = output.detach().cpu().numpy()
            out_label_ids = inputs["labels"].detach().cpu().numpy()
        else:
            pred_scores = np.append(pred_scores, output.detach().cpu().numpy(), axis=0)
            out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0)

    eval_loss = eval_loss / nb_eval_steps
    if args.output_mode in ["multi-label", "multi_label"]:
        result = metrics_multi_label(out_label_ids, pred_scores, threshold=0.5)
    elif args.output_mode in ["multi-class", "multi_class"]:
        result = metrics_multi_class(out_label_ids, pred_scores)
    elif args.output_mode == "regression":
        pass # to do
    elif args.output_mode in ["binary-class", "binary_class"]:
        result = metrics_binary(out_label_ids, pred_scores, threshold=0.5,
                                savepath=os.path.join(output_dir, "dev_confusion_matrix.png"))

    with open(os.path.join(output_dir, "dev_metrics.txt"), "w") as writer:
        logger.info("***** Eval Dev results {} *****".format(prefix))
        for key in sorted(result.keys()):
            logger.info("%s = %s", key, str(result[key]))
            writer.write("%s = %s\n" % (key, str(result[key])))

    logger.info("Dev metrics: ")
    logger.info(json.dumps(result, ensure_ascii=False))
    logger.info("")

    return result