Phase 2 secondary-prediction baseline: train + validate #2

New Issue

2026-07-08T13:52:17+02:00

lbogner commented

2026-07-08 13:52:17 +02:00

(Migrated from gitlab.etp.kit.edu)

Establish the Phase 2 baseline: train and validate the two-stage secondary-prediction model as currently implemented (Stage 1 DenoisingMLP + n_sec head, Stage 2 SecondaryDecoder), so we have a reference result to measure faster-eval alternatives against.

This is not a re-implementation — the code has landed on phase2-secondary-prediction. The remaining work is to produce the baseline numbers: train to convergence on the PbWO4 dataset, run step-level marginal + KL validation (primary 9D, n_sec, per-secondary energy/direction/species) and shower-level rollout validation via giant rollout against Geant4 truth, and record the resulting checkpoint as the reference that the WGAN and MoE alternatives are compared to.

Source: Jan & Tobias meeting 2026-07-07, action item "continue Phase 2 as-is to get a baseline."

Establish the **Phase 2 baseline**: train and validate the two-stage secondary-prediction model as currently implemented (Stage 1 `DenoisingMLP` + `n_sec` head, Stage 2 `SecondaryDecoder`), so we have a reference result to measure faster-eval alternatives against. This is not a re-implementation — the code has landed on `phase2-secondary-prediction`. The remaining work is to produce the baseline numbers: train to convergence on the PbWO4 dataset, run step-level marginal + KL validation (primary 9D, `n_sec`, per-secondary energy/direction/species) and shower-level rollout validation via `giant rollout` against Geant4 truth, and record the resulting checkpoint as the reference that the WGAN and MoE alternatives are compared to. Source: Jan & Tobias meeting 2026-07-07, action item "continue Phase 2 as-is to get a baseline."

lbogner (Migrated from gitlab.etp.kit.edu) closed this issue

2026-07-17 12:33:09 +02:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: lars/giant#2