def find_best_parametrization_fast()

in muss/fairseq/main.py [0:0]

• 14 lines of code
• 6 McCabe index (conditional complexity)

def find_best_parametrization_fast(exp_dir, preprocessors_kwargs, **kwargs):
    preprocessors_kwargs = preprocessors_kwargs.copy()  # We are going to modify it inplace
    preprocessors = get_preprocessors(preprocessors_kwargs)
    orig_sents, refs_sents = get_orig_and_refs_sents(**kwargs.get('evaluate_kwargs', {'test_set': 'asset_valid'}))
    features = defaultdict(list)
    for ref_sents in refs_sents:
        for orig_sent, ref_sent in zip(orig_sents, ref_sents):
            for preprocessor in preprocessors:
                if not hasattr(preprocessor, 'get_feature_value'):
                    continue
                features[preprocessor.__class__.__name__].append(preprocessor.get_feature_value(orig_sent, ref_sent))
    for preprocessor_name, preprocessor_features in features.items():
        preprocessors_kwargs[preprocessor_name]['target_ratio'] = np.mean(preprocessor_features)
    return preprocessors_kwargs