Repository navigation
Merged
Conversation
… A overview display
fix(forge): refresh current-source family readouts and restore Cohort A overview display
… controls Load the exact eight original discovery cards from scientific commit 4749b27 with per-case hashes and Git-source provenance. Preserve both frozen coordinator sources, hard-coded gates and evidence pins. Bounded optimizer binding controls explicitly use current seed 0 software discovery and establish no historical replay or qualification credit. Align active benchmarks/toy_audit/api_family_search.py host_recipe_options with current VectorFixture ownership: task resources, initial prior law and declared schedule budget only; rates, losses, penalties, betas and EMA remain recipe-owned except explicit caller overrides. Root must reseal publication source bindings for this active-source change. Validate 341 critic/generator/family-search software tests; no training campaign or generated publication change.
Predict source-bound preflight refusal for any hash mismatch without classifying unrelated API contract changes as a constructor rename. Keep frozen_restoration_source and its exact byte/hash/module checks unchanged; no production source, frozen task or evidence pin changes. Add controls for matching bytes, unrelated changed bytes, exact rename proof and extra changes inside a declared restoration module.
Move Gaussian1D task-owned prior kind, absolute sigma and standardization into explicit prior_options with fixed sigma_rel=0; retain the declared 256-by-2 MoG and public recipe defaults. Independently test initial rows, sample law, capacity and exact recipe identity across K3P and BISPF. No hidden trainer rate, beta, regularizer or EMA overrides are restored. Archive original model-toy api_contract.py and api_run.py bytes from origin bee9f06 with exact SHA256 and Git blob identities. Views controls verify those originals and require refusal of changed current source rather than rebinding immutable task hashes. Preserve all report/source pins and qualification outcomes. Full scalar/views/archive modules: 126 bounded software tests passed; no scientific campaign or publication regeneration.
Repair scientific contracts and publish evidence without formula substitutions
…lap-20261022 Explain all eleven recorded BISPF failures in one review
…-20261022 Test R1/R2 repairs and fix Forge execution defects
…20261109 Integrate DDGAN, VAEGAN and independent MisGAN source contracts
…ore subtraction, and harden lint runner
…ist-20261109 Shrink PyPI sdist, mask observed NaNs before subtraction, and harden lint runner
…ract Reject unsupported MisGAN model and precision settings
martyn
changed the title
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
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.
Release ParticleGAN 0.9.8 from
developtomaster.Highlights in 0.9.8
Tiny PyPI distributions (
.whland.tar.gzsdist): AddMANIFEST.inpruning non-package directories (tests/,reports/,benchmarks/,experiments/,docs/,examples/,ops/,scripts/, etc.), shrinkingparticlegan-0.9.8.tar.gzfrom1.7 MB(7.14 MBuncompressed) to~343 KiB(~373 KiBwheel) and enforcing< 600 KBsize and root allowlists in CI and packaging tests (Shrink PyPI sdist, mask observed NaNs before subtraction, and harden lint runner #538).Reviewed model objectives, pre-subtraction
NaNmasking, and top-level exports: Exportconditional_energy_score,joint_reconstruction_loss,masked_mixture_nll,NeuralComponentImputer,masked_quantile_transport_loss,EmulatedLinear, andemulate_linear_training_from top-levelparticlegan. Maskobservedviatorch.wherebefore subtraction inmasked_mixture_nll(Integrate DDGAN, VAEGAN and independent MisGAN source contracts #537, Shrink PyPI sdist, mask observed NaNs before subtraction, and harden lint runner #538).Formula-faithful recipe admission, independent critic histories, and provenance: Enforce formula-faithful admission across vector, BiGAN, DDGAN, VAE-GAN, and MisGAN hosts (including Repair scientific contracts and publish evidence without formula substitutions #535, Reject unsupported MisGAN model and precision settings #536, Integrate DDGAN, VAEGAN and independent MisGAN source contracts #537, Explain all eleven recorded BISPF failures in one review #473, Test R1/R2 repairs and fix Forge execution defects #474), separate faithful numerical
FAILfromBLOCKED/INCOMPLETE/UNMEASUREDoutcomes, and preserve byte-exact historical receipts with ancestral source-binding verification.Exact trainer checkpoint continuation with gradient response and completed-update boundaries.
Technique inventory: https://github.com/255BITS/ParticleGAN/blob/develop/reports/forge/technique-inventory.md