Time-Series Split Boundaries

Problem

Return expanding time-series split boundaries. For the i-th of n_splits folds, the test window ends at n - (n_splits - 1 - i) * test_size and is test_size points long, while everything before it is training. Each entry is [train_end, test_start, test_end] with train_end equal to test_start. Return [] unless n_splits > 0, test_size > 0, and n_splits * test_size < n.

Hint

Each fold trains only on the past and tests on the next fixed-size block of the future.

Test cases

case 1

in: [10,3,2]

expected: [[4,4,6],[6,6,8],[8,8,10]]

case 2

in: [6,2,2]

expected: [[2,2,4],[4,4,6]]

case 3

in: [5,5,1]

expected: []

case 4

in: [4,1,1]

expected: [[3,3,4]]

Starter codePython

def time_series_split_bounds(n, n_splits, test_size):
    # Your code here
    pass

4 test cases · run your code against them in the editor