Documentation about tscv HOT 3 CLOSED

The data set is assumed to be in time order, though an explicit datatime column is not required. It should work on lists, arrays, dataframes. If not, please report a bug.

The following example comes from my blog post:

import numpy as np
from tscv import gap_train_test_split
X, y = np.arange(20).reshape((10, 2)), np.arange(10)
X_train, X_test, y_train, y_test = gap_train_test_split(X, y, test_size=2, gap_size=2)

In the example, X and y are numpy arrays. They don't have data time information, but they represent time-ordered data.

from tscv.

Thank you for clarifying. I eventually figured it out when I saw a reference in the code comments.

I recommend you make it clear in your documentation.

Also, do you have a layout recommendations/guidance on how to set combination layouts of split/test size/gap size?

I made some guesses when using and in some cases got errors. I'm using GapWalkForward and given that it ignores data after the test set I'm trying to easy figure exactly what I'm using in the folds so I'm making best use of data.

Thank you so much for creating and publishing this package

from tscv.

The documentation part is on the roadmap (see the v0.1.0 milestone).

do you have a layout recommendations/guidance on how to set combination layouts of split/test size/gap size?

This topic is publication worthy. There is no single fixed rule that can handle all cases, and many heuristics are available for choice. I do have some professional insight on this issue, but they cannot be made clear within a couple of sentences. I can give you some quick and dirty advice though: try some small gap sizes and check whether your conclusion stays the same with and without the gaps (cf. stress test and scenario analysis).

from tscv.