python - GroupSplitShuffle 和 GroupKFolds 之间的区别

Question

正如标题所说，我想知道 sklearnGroupKFold和GroupShuffleSplit.

两者都为具有组 ID 的数据进行训练测试拆分，因此这些组不会在拆分中分离。我检查了每个函数的一个训练集/测试集，它们看起来都做了一个很好的分层，但如果有人能确认所有拆分都这样做，那就太好了。

我对两者进行了测试，分为 10 次：

gss = GroupShuffleSplit(n_splits=10, train_size=0.8, random_state=42)

 

for train_idx, test_idx in gss.split(X,y,groups):

    print("train:", train_idx, "test:", test_idx)

train: [ 1  2  3  4  5 11 12 13 14 15 16 17 19 20] test: [ 0  6  7  8  9 10 18]

train: [ 1  2  3  4  5  6  7  8  9 10 12 13 14 18 19 20] test: [ 0 11 15 16 17]

train: [ 0  1  3  4  5  6  7  8  9 10 12 13 14 18 19 20] test: [ 2 11 15 16 17]

train: [ 0  2  3  4 11 12 13 14 15 16 17 18 19 20] test: [ 1  5  6  7  8  9 10]

train: [ 0  1  3  4  5  6  7  8  9 10 11 15 16 17 19 20] test: [ 2 12 13 14 18]

train: [ 1  2  3  4  5  6  7  8  9 10 11 15 16 17 18] test: [ 0 12 13 14 19 20]

train: [ 0  1  2  3  4  6  7  8  9 10 11 12 13 14 15 16 17] test: [ 5 18 19 20]

train: [ 0  1  3  4  6  7  8  9 10 11 15 16 17 18 19 20] test: [ 2  5 12 13 14]

train: [ 0  1  3  4  5 12 13 14 15 16 17 18 19 20] test: [ 2  6  7  8  9 10 11]

train: [ 0  2  3  4  5 11 12 13 14 15 16 17 19 20] test: [ 1  6  7  8  9 10 18]

 

group_kfold = GroupKFold(n_splits=10)

 

for train_idx, test_idx in group_kfold.split(X,y,groups):

    print("train:", train_idx, "test:", test_idx)

train: [ 0  1  2  3  4  5 11 12 13 14 15 16 17 18 19 20] test: [ 6  7  8  9 10]

train: [ 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 18 19 20] test: [15 16 17]

train: [ 0  1  2  3  4  5  6  7  8  9 10 11 15 16 17 18 19 20] test: [12 13 14]

train: [ 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18] test: [19 20]

train: [ 0  1  2  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20] test: [3 4]

train: [ 1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 19 20] test: [ 0 18]

train: [ 0  1  2  3  4  5  6  7  8  9 10 12 13 14 15 16 17 18 19 20] test: [11]

train: [ 0  1  2  3  4  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20] test: [5]

train: [ 0  1  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20] test: [2]

train: [ 0  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20] test: [1]

标签： pythonscikit-learnsplitcross-validationk-fold

Answer 1

un-Group 版本的文档更清楚地说明了这一点。KFold 拆分为 k 个折叠，然后将它们集中到不同的训练/测试拆分中，而 ShuffleSplit 重复直接进行训练/测试拆分。特别是，每个样本在 KFold 中只测试一次，但可以在 ShuffleSplit 中测试零次或多次。