[BUG] RDST admissible sampling point method returns wrong index

[baraline]

Describe the bug

In RandomDilatedShapeletTransform the _get_admissible_sampling_point returns the wrong timestamp index.

Currently, the function returns:

return i, idx_choice

while it should return

return i, current_mask[i][idx_choice]

This results in some shapelets being sampled from the same point, that should have been banned from previous step, augmenting the number of duplicate shapelets that are useless for the computation, and reducing the number of "different" shapelets being sampled.

Performance analysis should be conducted before / after the fix to asses the impact.

Steps/Code to reproduce the bug

No response

Expected results

N/A

Actual results

N/A

Versions

No response

[baraline]

Results are somewhat equivalent on univariate (default train/test) Number of shapelet sampled is below the default (10k) for small datasets as expected, will push the fix.