eval_performance.py 文件源码

def evaluate(predictions, labels, threshold=0.4, multi_label=True):
    '''
        True Positive  :  Label : 1, Prediction : 1
        False Positive :  Label : 0, Prediction : 1
        False Negative :  Label : 0, Prediction : 0
        True Negative  :  Label : 1, Prediction : 0
        Precision      :  TP/(TP + FP)
        Recall         :  TP/(TP + FN)
        F Score        :  2.P.R/(P + R)
        Ranking Loss   :  The average number of label pairs that are incorrectly ordered given predictions
        Hammming Loss  :  The fraction of labels that are incorrectly predicted. (Hamming Distance between predictions and labels)
    '''
    assert predictions.shape == labels.shape, "Shapes: %s, %s" % (predictions.shape, labels.shape,)
    metrics = dict()
    if not multi_label:
        metrics['bae'] = BAE(labels, predictions)
        labels, predictions = np.argmax(labels, axis=1), np.argmax(predictions, axis=1)

        metrics['accuracy'] = accuracy_score(labels, predictions)
        metrics['micro_precision'], metrics['micro_recall'], metrics['micro_f1'], _ = \
            precision_recall_fscore_support(labels, predictions, average='micro')
        metrics['macro_precision'], metrics['macro_recall'], metrics['macro_f1'], metrics['coverage'], \
            metrics['average_precision'], metrics['ranking_loss'], metrics['pak'], metrics['hamming_loss'] \
            = 0, 0, 0, 0, 0, 0, 0, 0

    else:
        metrics['coverage'] = coverage_error(labels, predictions)
        metrics['average_precision'] = label_ranking_average_precision_score(labels, predictions)
        metrics['ranking_loss'] = label_ranking_loss(labels, predictions)

        for i in range(predictions.shape[0]):
            predictions[i, :][predictions[i, :] >= threshold] = 1
            predictions[i, :][predictions[i, :] < threshold] = 0

        metrics['bae'] = 0
        metrics['patk'] = patk(predictions, labels)
        metrics['micro_precision'], metrics['micro_recall'], metrics['micro_f1'], metrics['macro_precision'], \
            metrics['macro_recall'], metrics['macro_f1'] = bipartition_scores(labels, predictions)
    return metrics