Review v1.5.0#45
Conversation
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
Fix wavelet norm
Updated the CHANGELOG to include new features and fixes for version 1.5.0.
Update CHANGELOG for version 1.5.0
| r2 = r**2 | ||
| support_r = (r > 0.0) & (r < 1.0) | ||
| support_r = (r >= 0.0) & (r < 1.0) | ||
| r2 *= width_factor # optimized parameter (for L=4) to follow at best Littlewood Paley condition, see numerical integration in the notebook "BS_wavelet_kernel.ipynb" |
There was a problem hiding this comment.
Shouldn't this width factor also be applied to the support's indicator function? Otherwise, the radial bump is no longer continuous, and its value blows up toward infinity on both sides of the bump:
There was a problem hiding this comment.
Actually the line denom = (1.0 - r2).clamp_min(eps) which comes just after puts the high valued segments (>1) to exp(-r2 / eps) (almost 0) so there is no discontinuity. But for readability I can limit also the r support to [0, width_factor**-.5[.
There was a problem hiding this comment.
@dtibi69 I limited the r support to [0, width_factor**-.5[ here: https://github.com/STL-Proj/STL-Dev/pull/46/changes#diff-cc3e4a109ef54880c3ed70bb9dbd0ccb68bb00d1e34890e9fad079afb1231211
2 participants
No description provided.