Skip to content

CHANGELOG.md

Latest commit

 

History

History
43 lines (33 loc) · 1.89 KB

CHANGELOG.md

File metadata and controls

43 lines (33 loc) · 1.89 KB
Raw

Changelog

All notable changes to HGraphML will be documented here.

This project follows semantic-versioning vocabulary, but early releases should be read as research milestones rather than stability promises.

Unreleased

Tensorization compatibility bridge.

  • Updated the state-collapser dependency pin to the public state_collapser v0.7.0 tag.
  • Added build_encoding_registry(...), a thin HGraphML adapter helper around upstream EncodingRegistry.from_tower(...).
  • Added compatibility tests covering registry construction, JSON-safe registry metadata, base state and edge encodability, state-cell and action-cell encodability, and HGraphML node/edge fiber coverage.
  • Documented that this is shared tower encoding compatibility for future tensorization work, not full graph-message tensorization, RL transition tensorization, or benchmark-backed speed-up.

0.1.0 - Research Viability Release

Initial lightweight public research release.

This release establishes the first executable bridge from state_collapser quotient towers into trainable graph message passing:

  • Added the TensorGraph graph surface.
  • Added a direct adapter from known graph data into state_collapser towers.
  • Added node-fiber and edge-fiber readouts from tower tiers.
  • Added deterministic uniform and fiber-normalized lifts.
  • Added a learned PyTorch lift.
  • Added message containers, pooling, edge-message MLPs, and readout helpers.
  • Added collapse_messages(...) as the package-native orchestration call.
  • Added a small supervised train-step helper and runnable learned-lift demo.
  • Added tests, type checking, Ruff linting, build metadata, README, usage docs, API notes, design notes, and contributor guidance.

This release does not claim graph-ML speed-up. It demonstrates trainable quotient-tower-backed message passing and sets up the benchmarking work needed before performance claims are appropriate.