You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
the LGCD solver (which is the default for BatchCDL) seems to run into some issues when trying on new data larger than a certain size. Specifically, when I use the transform function, it freezes above a certain size threshold but works perfectly below it. None of the other solvers seem to have the issue, but they also dont seem to work as well.
The text was updated successfully, but these errors were encountered:
and to look at each segment. seg_bounds = [0, n_times_seg]
seg_bounds[0] += n_times_seg
seg_bounds[1] += n_times_seg
Most of the times here, n_times_seg will be fixed to n_times_seg = 2 * n_times_atom + 1. When n_times_valid is large enough however, n_seg becomes larger than n_times_seg.
This is an issue because here, there are according to each segment, we should have n_seg blocks of size n_times_seg.
However, the n_seg and n_seg_times are defined as if we are solving for n_times_seg blocks of size n_seg.
Specifically, we have n_seg*(n_times_seg-1) < n_times_valid < n_seg*n_times_seg instead of having (n_seg-1)n_times_seg < n_times_valid < n_segn_times_seg.
As a result, for larger datasets, the program will often freeze since too many blocks are defined that are never iterated over and therefore never "solved".
The two fixes for this are:
Make it so that the functions solves for n_times_seg blocks of size n_seg. For example, having instead seg_bounds = [0, n_seg]
seg_bounds[0] += n_seg
seg_bounds[1] += n_seg
Make it so that the number of blocks is properly defined. n_times_seg = (2 * n_times_atom) + 1
n_seg = n_times_valid // n_times_seg + 1
I personally did option 2, though I'm not sure which one is more consistent with your paper. Let me know if there was a mistake here.
the LGCD solver (which is the default for BatchCDL) seems to run into some issues when trying on new data larger than a certain size. Specifically, when I use the transform function, it freezes above a certain size threshold but works perfectly below it. None of the other solvers seem to have the issue, but they also dont seem to work as well.
The text was updated successfully, but these errors were encountered: