我应该使用什么而不是引导程序

Question

当我运行这段代码时：

from sklearn import cross_validation
bs = cross_validation.Bootstrap(9, random_state=0)

我收到此弃用警告：

C:Anacondaenvsp33libsite-packagessklearncross_validation.py:684: DeprecationWarning: Bootstrap will no longer be supported as a cross-validation method as of version 0.15 and will be removed in 0.17
  "will be removed in 0.17", DeprecationWarning)

我应该使用什么而不是引导程序？

Answer 1

来自scikit-learn 0.15发行说明，在"API更改摘要"下

• cross_validation.Bootstrap已弃用。 建议改用cross_validation.KFold或cross_validation.ShuffleSplit。

从源代码本身来看：

# See, e.g., http://youtu.be/BzHz0J9a6k0?t=9m38s for a motivation
# behind this deprecation
warnings.warn("Bootstrap will no longer be supported as a " +
              "cross-validation method as of version 0.15 and " +
              "will be removed in 0.17", DeprecationWarning)

Answer 2

您可以使用

BaggingClassifier：

bag = BaggingClassifier(base_estimator=your_estimator, 
                        n_estimators=100,
                        max_samples=1.0,
                        bootstrap=True,
                        n_jobs=-1)
bag.fit(X, y)
recalls = []
for estimator, samples in zip(bag.estimators_, bag.estimators_samples_):
    # compute predictions on out-of-bag samples
    mask = ~samples
    y_pred = estimator.predict(X[mask])
    # compute some statistic
    recalls.append(recall(y[mask], y_pred))
# Do something with stats, e.g. find confidence interval
print(np.percentile(recalls, [2.5, 97.5]))

Answer 3

我刚刚遇到了这个问题，我找到的解决方案（从scikit-learn 0.24开始）是使用重新采样实用程序。

from sklearn.utils import resample

这将在每次调用时生成 1 个引导步骤，使用默认参数进行采样并替换。

https://scikit-learn.org/stable/modules/generated/sklearn.utils.resample.html