def evaluate()

in evaluation/evaluate_hf.py [0:0]

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

def evaluate(benchmark: str, dataset_id: str, dataset_config: str = None, dataset_split: str = "test", dataset_col: str = "pred", samples: list=None, max_num_samples=None):
    samples = load_dataset(dataset_id, name=dataset_config, split=dataset_split)
    if "idx" not in samples.column_names:
        samples = samples.map(lambda x, idx: {"idx": idx}, with_indices=True)
        
    if max_num_samples:
        print(f"max_num_samples: {max_num_samples} / {len(samples)}")
        samples = samples[:max_num_samples]


    def parse_gt(x):
        x['gt_cot'], x['gt'] = parse_ground_truth(x, benchmark)
        return x
    samples = samples.map(parse_gt, desc="Parsing ground truth", num_proc=12, load_from_cache_file=False)
    samples = samples.map(extract_answer_map, fn_kwargs={"data_name": benchmark, "col": dataset_col}, desc="Parsing predictions", num_proc=12, load_from_cache_file=False)
    params = [(idx, pred, gt) for idx, pred, gt in zip(samples['idx'], samples['pred'], samples['gt'])]

    scores = []
    timeout_cnt = 0 

    with ProcessPool(max_workers=8) as pool:
        future = pool.map(math_equal_process, params, timeout=3)
        iterator = future.result()
        with tqdm(total=len(samples), desc="Evaluate") as progress_bar:
            while True:
                try:
                    result = next(iterator)
                    scores.append(result)
                except StopIteration:
                    break
                except TimeoutError as error:
                    print(error)
                    scores.append(False)
                    timeout_cnt += 1
                except Exception as error:
                    print(error.traceback)
                    exit()
                progress_bar.update(1) 

    mean_score = np.mean(scores) * 100

    result_json = {
        "num_samples": len(samples),
        "num_scores": len(scores),
        "timeout_samples": timeout_cnt,
        "acc": mean_score
    }

    print(result_json)
    return samples, result_json