v0.5.0

What's Changed

• Closed-form CRPS for Gaussian and posterior predictive distributions (0.5.0) by @debrevitatevitae in #149

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

• Posterior validation scores fixed draws under a constant default key by @debrevitatevitae in #141

Full Changelog: v0.3.0...v0.4.0

v0.3.0

What's Changed

• Performance: compile-stable, measured training and prediction (Tier 1) by @debrevitatevitae in #134

Full Changelog: v0.2.0...v0.3.0

v0.2.0

stagereg 0.2.0

This release adds MC-Dropout as a third posterior inference method, alongside BayesByBackprop and PreconditionedSGLD. It also lets you infer a posterior over only part of a network, and adds tooling for checkpoints and evaluation.

Highlights

• MCDropout and MCDropoutPosterior: fine-tunes the inferred subnetwork with dropout turned on, and draws one dropout mask set per key. It supports early stopping and checkpoints. The XSin posterior example now demonstrates it (#92, #119).
• Inferring part of a network: the new PosteriorStage.inferred_subnetwork filter picks the subnetwork to infer, and the rest of the network stays fixed at its trained-mean values. The default still infers the whole network (#105).
• Diagnostics from inference methods: InferenceMethod.diagnostics names extra update outputs, which are recorded as posterior/train/<key> epoch means (#106).
• expected_sample_crps: computes the exact expected CRPS without a grid, batched over scoring units.
• CachedCheckpointStore: caches every save in memory and passes only the saves you select to a persistent store. Ready-made rules for which saves to keep: finalized_only, every_nth_save(n) and always.
• MeanStage and GammaVarianceStage now save a finalized checkpoint even without an early stopper, so a later process can restore them.

⚠️ Breaking changes

• Keyed posterior functions: mean_function and log_likelihood now take a PRNG key, on PosteriorProblem and on both posterior classes. A deterministic network gives the same results as before (#92).
• InferenceMethod needs a diagnostics property: a custom method can declare () to keep its current behaviour (#106).
• Posterior problems and priors cover only the inferred subnetwork. load_posterior refuses a posterior whose parameter tree doesn't match that subnetwork (#105).
• Training-state fields cleaned up:
  • TrainingState.stage is replaced by the active_stage field.
  • Checkpoint.rng_state is removed; the key is in Checkpoint.state.rng_state.
  • outer_iteration, data_fingerprint and TrainingEvent.iteration are removed.

See CHANGELOG.md for the full list.

v0.1.0

What's Changed

• Add backend-neutral early stopping and optional JAX training backend by @debrevitatevitae in #1
• chore: add CI workflow with lint and test jobs by @debrevitatevitae with @Copilot in #2
• Add structured issue intake and contributor workflow guide by @debrevitatevitae with @Copilot in #4
• feat: add distribution-valued Gaussian head and MVE training stage by @debrevitatevitae in #3
• Add composable metric registration for JAX training/validation loops by @debrevitatevitae with @Copilot in #6
• Add mean and Gamma variance training stages by @debrevitatevitae in #8
• Audit: property-based testing by @debrevitatevitae in #9
• Rearrange examples by @debrevitatevitae in #10
• feat: add CMAPSS FD001 data exploration by @debrevitatevitae in #12
• feat: RUL curve plots with 95% predictive intervals for the CMAPSS example by @debrevitatevitae in #23
• feat: tracking sink, shared in-memory checkpoint store, README example paths by @debrevitatevitae in #33
• feat!: stage/split/metric naming scheme by @debrevitatevitae in #50
• docs: user documentation site (MkDocs Material + mkdocstrings) by @debrevitatevitae in #69
• docs: document the attributes of every public dataclass by @debrevitatevitae in #75
• Stages restore and finalize their own checkpoints under stage-scoped keys by @debrevitatevitae in #80
• Posterior stage (VeBNN Step 3) with pluggable inference methods by @debrevitatevitae in #90
• xsin example: widen epistemic bands in extrapolation by @debrevitatevitae in #93
• docs: motivate stage-based training on the docs homepage by @debrevitatevitae in #97
• Rename to stagereg and release 0.1.0 by @debrevitatevitae in #104

New Contributors

• @debrevitatevitae with @Copilot made their first contribution in #2

Full Changelog: https://github.com/GroupiSP/stagereg/commits/v0.1.0