Threshold_selection.py

def threshold_search(y_true, y_proba):
    best_threshold = 0
    best_score = 0
    for threshold in tqdm([i * 0.01 for i in range(100)]):
        score = K.eval(matthews_correlation(y_true.astype(np.float64), (y_proba > threshold).astype(np.float64)))
        if score > best_score:
            best_threshold = threshold
            best_score = score
    search_result = {'threshold': best_threshold, 'matthews_correlation': best_score}
    return search_result

ok