GapKFold
This page describes K-fold and how to use gaps with it for time series.
The cross-validation known as K-Fold may be the most wildly used cross-validation method in machine learning. It devides the sample into \(K\) folds and each time uses \(K-1\) folds as training set and the remaining 1 fold as test set. There are \(K\) different test sets in all. To ensure that the every group of \(K\) data points has equal chance of being choosed, the sample is shuffled before hand. Since \(K \ll C_n^p\), it be more efficient than leave-p-out.
For time series cross-validation, the shuffle procedure is inappropriate. We simply add gaps on both sides of the test fold. The gaps are borrowed from neighouring training folds. That is the neighouring training folds lose some sample points.
The following code snippet produces the cross-validation setup of the above image. Notice how the gap size before the test set can be different from the gap coming after:
>>> from tscv import GapKFold
>>> cv = GapKFold(n_splits=5, gap_before=2, gap_after=1)
>>> for train, test in cv.split(range(10)):
... print("train:", train, "test:", test)
...
train: [3 4 5 6 7 8 9] test: [0 1]
train: [5 6 7 8 9] test: [2 3]
train: [0 1 7 8 9] test: [4 5]
train: [0 1 2 3 9] test: [6 7]
train: [0 1 2 3 4 5] test: [8 9]
In the code sample, GapKFold is a class provided by this package.
It has a split method, which takes in the whole data set and produce
the training and test sets indices.