Repository navigation
Fix minor numerics: rotation precision, Cn invert pivots, Octa LM acceptance - #57
Merged
Merged
Conversation
…eptance * rotation.rotate_spherical_function: interpolate in the input precision instead of casting the float64 sampling grid and signal to float32 before grid_sample (identity rotation is now exact to 1e-12 in float64; half and bfloat16 inputs are promoted to float32 as before). * CnonCn.invert: document that Algorithm 1 divides by every fhat[k], k<=n-2, raise ValueError on a near-zero intermediate pivot instead of producing nan, and reject non-finite beta up front. * OctaonOcta._lm_step: compare each damping retry against the best candidate so far rather than the starting residual (a less damped retry that beat the start but not the accepted step used to overwrite it), and use a per-sample schedule (grow mu until first acceptance, shrink while improving, stop on first stall). Seeded A/B over 320 random float64 signals: slow-converging (>1e-6) 15 -> 5, gross failures 3 -> 1, median residual 7e-12 -> 3e-16. Tests: identity-rotation exactness (float64/float32) and R/R^T roundtrip; Cn zero-pivot and non-finite-beta errors; LM step never returns a point worse than the first accepted candidate (seed 35 was a concrete failure).
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #57 +/- ##
==========================================
+ Coverage 95.46% 95.48% +0.01%
==========================================
Files 11 11
Lines 2227 2236 +9
==========================================
+ Hits 2126 2135 +9
Misses 101 101 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
johmathe
added a commit
that referenced
this pull request
Oct 14, 2026
* Release v0.4.0 Bump version to 0.4.0 and record the feature-layout changes in the SO3onS2, DnonDn and SO2onDisk class docstrings (versionchanged notes). Breaking: output_size / layout changes for SO3onS2 (lmax >= 3), DnonDn (odd n >= 5) and the default SO2onDisk truncation. * Raise coverage to ~98.6%: drop dead code, test dense-CG cache and graph wiring Dead code removed (never reachable from the library): * _bessel: scalar _jn_scalar/_djn_scalar/_bisect_newton (only tests used them; the batch solver is the single root path) and the unreachable num_brackets <= 0 branch in compute_all_bessel_roots. * dn_on_dn: n2d == 0 / n3 == 0 branches (n >= 3 implies n2d >= 1). * so3_on_s2: empty-slice guard in the power loop of _ensure_precomputed_on_device (empty entries are never flagged power). Bug found by the new tests: _load_or_compute_reduced_cg did not catch pickle.UnpicklingError, so a corrupt reduced-CG cache file crashed module construction instead of being recomputed. New tests: package version fallback; _bessel_jn_multi(()) and order-0 batch roots; _save_cache write failure; triangle-violating sparse CG entries (empty tables, exact zeros in forward); SO3onS2 dense tables vs sparse forward (incl. CG power); reduced-CG disk cache compute/load/ corrupt/write-failure paths; lazy sparse gather-table construction; CUDA graph capture wiring with torch.cuda stream/graph APIs stubbed on CPU. Remaining uncovered lines are defensive numerical fallbacks in the D_n / octahedral CG constructions and the CUDA dispatch in SO3onS2.forward.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Three small numerical fixes, each TDD'd (test red on
main, green here).rotation.rotate_spherical_functionInterpolates in the input precision instead of casting the float64 sampling grid and signal to float32 before
grid_sample. Identity rotation is now exact to 1e-12 in float64 (was ~1e-5). half/bfloat16 are still promoted to float32.CnonCn.invertAlgorithm 1 divides by every
fhat[k], k ≤ n−2, not justfhat[0]andfhat[1]. Documented; a near-zero intermediate pivot now raisesValueErrorinstead of propagatingnan, and non-finite β is rejected up front.OctaonOcta._lm_stepEach damping retry was compared against the loss at the starting point, so a later less-damped retry that beat the start but not the already-accepted step overwrote it (seed 35 in the test: returned residual ~19 above the accepted one). Now compares against the best so far with a per-sample schedule (grow μ until first acceptance, shrink while improving, stop on first stall).
Seeded A/B over 320 random float64 signals: slow-converging (>1e-6) 15 → 5, gross failures 3 → 1, median residual 7e-12 → 3e-16; float32 64/64 clean.
Part of the v0.4.0 stack (E).