lars/giant
1
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects Releases Wiki Activity

Add class-balanced secondary particle-type loss (gitea #44) #70

Merged
lars merged 1 commits from fix/issue-44 into master 2026-08-18 10:41:45 +02:00
Conversation 0 Commits 1 Files Changed 11
+360 -22
lars commented 2026-08-18 10:24:00 +02:00
Owner
Copy Link
Copy Source

The v0.3.0 pivot exists because the 2026-08-03 WGAN rollout benchmark
produced zero photon secondaries and ~4M hallucinated antineutrinos —
even with a correctly-sized top-N species vocabulary (gitea #29), plain
cross-entropy over a class distribution spanning orders of magnitude
still under-predicts rare-but-physical species.

stage2_model.particle_type.class_weighting = "none" | "inverse_freq"
(default "none", fully back-compat) weights the stage-2 type head's CE
loss (FlowDDPMStageTrainer._type_loss) by inverse class frequency,
normalized to mean 1 so switching it on doesn't rescale the type loss
against particle_type.lambda / the generator loss it's summed with.

The per-class counts the weighting needs don't already exist despite the
issue's premise: _topn_plus_other_map (giant/data/loader.py) previously
kept counts only for keys folded into "other", dropping the kept classes'
counts on the floor. TopNMap now carries class_counts (index -> count),
round-tripped through the setup-cache sidecar (format version bumped
3->4, since existing sidecars have none) and through checkpoints
(tolerantly — a pre-#44 checkpoint decodes to {}, since only training-time
loss weighting reads it, not inference).

Decisions made during planning (with the user): dropped "effective_num"
from the issue's proposed three-way enum (no beta hyperparameter to
design around) — final domain is "none" | "inverse_freq". Weights are
mean-1-normalized. validate_config rejects class_weighting != "none"
combined with particle_type.target != "onehot" or
stage2_model.generator == "wgan" (both have no class CE to weight),
following the #28/#30 dead-key-must-not-go-silent convention. Branch
fix/issue-44 off master.

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com

The v0.3.0 pivot exists because the 2026-08-03 WGAN rollout benchmark produced zero photon secondaries and ~4M hallucinated antineutrinos — even with a correctly-sized top-N species vocabulary (gitea #29), plain cross-entropy over a class distribution spanning orders of magnitude still under-predicts rare-but-physical species. stage2_model.particle_type.class_weighting = "none" | "inverse_freq" (default "none", fully back-compat) weights the stage-2 type head's CE loss (FlowDDPMStageTrainer._type_loss) by inverse class frequency, normalized to mean 1 so switching it on doesn't rescale the type loss against particle_type.lambda / the generator loss it's summed with. The per-class counts the weighting needs don't already exist despite the issue's premise: _topn_plus_other_map (giant/data/loader.py) previously kept counts only for keys folded into "other", dropping the kept classes' counts on the floor. TopNMap now carries class_counts (index -> count), round-tripped through the setup-cache sidecar (format version bumped 3->4, since existing sidecars have none) and through checkpoints (tolerantly — a pre-#44 checkpoint decodes to {}, since only training-time loss weighting reads it, not inference). Decisions made during planning (with the user): dropped "effective_num" from the issue's proposed three-way enum (no beta hyperparameter to design around) — final domain is "none" | "inverse_freq". Weights are mean-1-normalized. validate_config rejects class_weighting != "none" combined with particle_type.target != "onehot" or stage2_model.generator == "wgan" (both have no class CE to weight), following the #28/#30 dead-key-must-not-go-silent convention. Branch fix/issue-44 off master. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
lars added 1 commit 2026-08-18 10:24:01 +02:00
Add class-balanced secondary particle-type loss (gitea #44)
CI / Lint (ruff check) (push) Successful in 36s
Details
CI / Format (ruff format) (push) Successful in 38s
Details
CI / Sync project version with tag (push) Has been skipped
Details
CI / Type check (ty) (push) Successful in 38s
Details
CI / Format (ruff format) (pull_request) Successful in 43s
Details
CI / Lint (ruff check) (pull_request) Successful in 45s
Details
CI / Sync project version with tag (pull_request) Has been skipped
Details
CI / Type check (ty) (pull_request) Successful in 47s
Details
CI / Tests (push) Successful in 5m20s
Details
CI / Tests (pull_request) Successful in 4m50s
Details
fce47b128c 
The v0.3.0 pivot exists because the 2026-08-03 WGAN rollout benchmark
produced zero photon secondaries and ~4M hallucinated antineutrinos —
even with a correctly-sized top-N species vocabulary (gitea #29), plain
cross-entropy over a class distribution spanning orders of magnitude
still under-predicts rare-but-physical species.

stage2_model.particle_type.class_weighting = "none" | "inverse_freq"
(default "none", fully back-compat) weights the stage-2 type head's CE
loss (FlowDDPMStageTrainer._type_loss) by inverse class frequency,
normalized to mean 1 so switching it on doesn't rescale the type loss
against particle_type.lambda / the generator loss it's summed with.

The per-class counts the weighting needs don't already exist despite the
issue's premise: _topn_plus_other_map (giant/data/loader.py) previously
kept counts only for keys folded into "other", dropping the kept classes'
counts on the floor. TopNMap now carries class_counts (index -> count),
round-tripped through the setup-cache sidecar (format version bumped
3->4, since existing sidecars have none) and through checkpoints
(tolerantly — a pre-#44 checkpoint decodes to {}, since only training-time
loss weighting reads it, not inference).

Decisions made during planning (with the user): dropped "effective_num"
from the issue's proposed three-way enum (no beta hyperparameter to
design around) — final domain is "none" | "inverse_freq". Weights are
mean-1-normalized. validate_config rejects class_weighting != "none"
combined with particle_type.target != "onehot" or
stage2_model.generator == "wgan" (both have no class CE to weight),
following the #28/#30 dead-key-must-not-go-silent convention. Branch
fix/issue-44 off master.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
lars merged commit c09ebd2410 into master 2026-08-18 10:41:45 +02:00
lars deleted branch fix/issue-44 2026-08-18 10:41:51 +02:00
Sign in to join this conversation.
Reviewers
No Reviewers
Labels
Clear labels
architecture
bug
Silent/wrong behavior, no crash
 experiment
feature
New feature matching the v0.3 redesign
 investigation
modularity
Making the network genuinely pluggable
 performance
phase2
validation
wontfix
Not being fixed right now
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
lars (Lars Bogner)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: lars/giant#70
Delete Branch "fix/issue-44"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?