Class-balanced secondary particle-type loss #44

New Issue

2026-08-13T15:08:03+02:00

lars commented

2026-08-13 15:08:03 +02:00

Complementary to the species-vocabulary-sizing issue (#29). With a top-16
species vocabulary over a distribution with a long tail, plain cross-entropy
will still under-predict rare-but-physical species — which is the observed
failure (zero photons). The class frequencies are already collected by
build_pdg_topn_map_from_files, so
particle_type.class_weighting = "none" | "inverse_freq" | "effective_num" is
close to free, and directly targets the metric the whole redesign is judged on.

Migrated from issues.md (v0.3.0 branch review, 2026-08-13), Issue 19.

Complementary to the species-vocabulary-sizing issue (#29). With a top-16 species vocabulary over a distribution with a long tail, plain cross-entropy will still under-predict rare-but-physical species — which is the observed failure (zero photons). The class frequencies are already collected by `build_pdg_topn_map_from_files`, so `particle_type.class_weighting = "none" | "inverse_freq" | "effective_num"` is close to free, and directly targets the metric the whole redesign is judged on. --- Migrated from `issues.md` (v0.3.0 branch review, 2026-08-13), Issue 19.

lars added the feature label 2026-08-13 15:08:03 +02:00

lars referenced this issue

2026-08-13 16:11:29 +02:00

Secondary species vocabulary is accidentally sized by conditioning.particle.emb_dim #29

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: lars/giant#44