Python 得到最坏的多类分类预测类

我正在研究一个多类分类问题，我有很多不同的类（50+）

问题是，我想突出显示预测最差的类（例如，在混淆矩阵或其他什么中），以便在分类器中进行进一步调整

我的预测和测试数据保存在列表中（来自sklearn的小示例）：

如何从矩阵中获取本例中的类1？在那里，预测是完全错误的。有没有一种方法可以根据类的真实积极预测对类进行排序？

您可以使用scikit learn，它将返回一个字典，其中包含精度、召回率和F分数。然后，您可以按排序方式打印字典，以便轻松查看最坏的预测类

#prints classification_report     
print(classification_report(y_true, y_pred)

#returns a dict, which you can easily sort by prediction
report = classification_report(y_true, y_pred, output_dict=True)

---

您可以为此使用一个简单的函数：

def print_class_accuracies(confusion_matrix):
    # get the number of occurrences for each class
    counts = {cl: y_true.count(cl) for cl in set(y_true)}
    # extract the diagonal values (true positives)
    tps = dict(enumerate(conf.diagonal()))
    # Get the accuracy for each class, preventing ZeroDivisionErrors
    pred_accuracy = {cl: tps[cl]/counts.get(cl, 1) for cl in tps}
    # Get a ranking, worst accuracies are first/lowest
    ranking = sorted([(acc,cl) for cl, acc in pred_accuracy.items()])
    # Pretty print it
    for acc, cl in ranking:
        print(f"Class {cl}: accuracy: {acc:.2f}")

def print_class_accuracies(confusion_matrix):
    # get the number of occurrences for each class
    counts = {cl: y_true.count(cl) for cl in set(y_true)}
    # extract the diagonal values (true positives)
    tps = dict(enumerate(conf.diagonal()))
    # Get the accuracy for each class, preventing ZeroDivisionErrors
    pred_accuracy = {cl: tps[cl]/counts.get(cl, 1) for cl in tps}
    # Get a ranking, worst accuracies are first/lowest
    ranking = sorted([(acc,cl) for cl, acc in pred_accuracy.items()])
    # Pretty print it
    for acc, cl in ranking:
        print(f"Class {cl}: accuracy: {acc:.2f}")