使用RandomForest进行K折交叉验证

Question

我目前正在尝试使用RandomForest进行预测，同时还使用k倍交叉验证来最小化min_samples_leaf的交叉验证错误。我目前无法设置代码，因为进入train_x = x[train_index]时我一直遇到错误。我收到的错误显示在下面。

from sklearn import model_selection
kf = model_selection.KFold(n_splits=5)

x = train
y = test

for m in range(0, 10): # vary min_samples_leaf

    dtr = ensemble.RandomForestRegressor(n_estimators = 15, min_samples_leaf = m, max_features = 10, criterion = 'mse')

    for train_index, test_index in kf.split(x):
        print("TRAIN:", train_index, "TEST:", test_index)
        train_x = x[train_index]
        train_y = y[test_index]
        regr = dtr.fit(train_x, train_y)

KeyError：

None of [Int64Index([15546, 15547, 15548, 15549, 15550, 15551, 15552, 15553, 15554,\n            15555,\n            ...\n            77718, 77719, 77720, 77721, 77722, 77723, 77724, 77725, 77726,\n            77727],\n           dtype='int64', length=62182)] are in the [columns]

Answer 1

您有从kf.split（）中提取的全部值，而您要调用x [train_index]的train_index不仅在数组x中。

代码似乎正确，所以我怀疑“ train”（当然还有“ x”）中的数据格式有问题吗？

错误表明您的Int64Index类型（索引IIRC的熊猫类型）的值大于x的值（最大长度为62182），因此原始数据肯定存在问题。