Skip to content

CHANGELOG.md

Latest commit

 

History

History
182 lines (117 loc) · 6.09 KB

CHANGELOG.md

File metadata and controls

182 lines (117 loc) · 6.09 KB

CHANGELOG

All notable changes to this project will be documented in this file.

This project adheres to Semantic Versioning and uses Conventional Commits.

Going forward, this file is updated automatically by cz bump on each release.

v1.8.0 (2026-05-24)

Feat

  • stable/experimental model split, save/load, and bug fixes
  • models: add save/load, fix get_params and LSS family pickling
  • stable and experimental model separation

Fix

  • duplicate entry in docs for model classes
  • test case fixed for windows
  • suppress family hparam warning
  • tabr: cast candidate_y to float in regression/LSS label encoder path
  • tabr: use regression label encoder for LSS by forwarding lss flag
  • types: guard task_model None, use getattr for get_params, annotate test kwargs
  • lint: drop unused unpacked variables and dead code
  • tabtrans: add dedicated LayerNorm for numerical features instead of reusing encoder norm
  • ndtf: correct ensemble aggregation for multi-class and LSS outputs
  • added faiss-cpu
  • add delu dependency for TabR
  • DefaultTabRConfig export
  • import error after experimental namespace changes
  • resolve stale site-packages

v1.7.0 (2026-05-14)

Documentation

  • Added Developer Guide section with dedicated pages for contributing, the release process, model promotion policy (experimental → stable), and a SPEC 0–aligned support matrix
  • Added Getting Started section: new Overview page with full model reference table, Installation guide (prerequisites, PyPI, source, optional Mamba CUDA kernels), and Key Concepts page covering the sklearn API, task variants, config system, and preprocessing
  • Rewrote all three example pages (classification, regression, distributional regression) as narrative tutorials with runnable code and "using your own data" sections
  • Added llms.txt index for LLM tool discovery
  • Switched to sphinx_design for cards and grid layouts; added custom CSS with JetBrains Mono font and brand colour palette
  • Added pygments_style = "friendly" and dark-mode code block colours via CSS
  • Updated release workflow documentation; fixed Mermaid diagram node line breaks

CI

  • Dropped Python 3.14 from the test matrix (scipy wheels are not yet available); ceiling reverted to <3.14 in pyproject.toml
  • Fixed torch upper bound from <=2.7.0 to <2.8.0 to allow patch releases
  • Replaced tomllib with poetry version --short for Python 3.10 compatibility in CI scripts
  • Added Lightning log filtering to reduce test output noise
  • Regenerated poetry.lock after dependency constraint changes

Bug Fixes

  • Fixed Pyright type errors in test fixtures (DataFrame construction and unused variable bindings)

v1.7.0rc2 (2026-05-09)

Documentation

  • Updated and fixed release workflow documentation
  • Applied doc theme and header size style refinements

CI

  • Fixed tomllib usage replaced with poetry version --short for Python 3.10 compatibility

v1.7.0rc1 (2026-05-08)

Documentation

  • Added Developer Guide section with dedicated pages for contributing, the release process, model promotion policy (experimental → stable), and a SPEC 0–aligned support matrix
  • Added Getting Started section: new Overview page with full model reference table, Installation guide (prerequisites, PyPI, source, optional Mamba CUDA kernels), and Key Concepts page covering the sklearn API, task variants, config system, and preprocessing
  • Rewrote all three example pages (classification, regression, distributional regression) as narrative tutorials with runnable code and "using your own data" sections
  • Added llms.txt index for LLM tool discovery
  • Switched to sphinx_design for cards and grid layouts; added custom CSS with JetBrains Mono font and brand colour palette
  • Added pygments_style = "friendly" and dark-mode code block colours via CSS

CI

  • Dropped Python 3.14 from the test matrix (scipy wheels are not yet available); ceiling reverted to <3.14 in pyproject.toml
  • Fixed torch upper bound from <=2.7.0 to <2.8.0 to allow patch releases
  • Added Lightning log filtering to reduce test output noise
  • Regenerated poetry.lock after dependency constraint changes

Bug Fixes

  • Fixed Pyright type errors in test fixtures (DataFrame construction and unused variable bindings)

v1.6.1 (2025-04-26)

Changes

  • Renamed package from mambular / deeptabular to deeptab
  • Dynamic versioning: version is now sourced from pyproject.toml via importlib.metadata; removed __version__.py
  • CI rework: split into lint, typecheck, build, and test jobs; manual tagging with OIDC PyPI publishing; removed semantic-release automation

v1.5.0 (2025-04-14)

Changes

  • Moved preprocessing to the pretab package; removed bundled preprocessor
  • Added TabR model integration
  • Fixed LSS bug affecting distributional output
  • Updated docstrings for documentation generation compatibility

v1.4.0 (2025-03-24)

Features

  • Added ModernNCA model
  • Added training candidates support during prediction and validation in the lightning module

v1.3.2 (2025-03-19)

Bug Fixes

  • Fixed num_classes argument for binary classification
  • Fixed version info extraction

v1.3.1 (2025-03-17)

Features

  • Added Tangos model (classifier, regressor, and distributional variants)

v1.3.0 (2025-03-13)

Features

  • Added AutoInt model
  • Added Trompt model
  • Added ENode (embedding oblivious trees) model
  • Fixed preprocessor bug causing ValueError: not enough values to unpack

v1.2.0 (2025-02-17)

Features

  • Added BaseConfig parent class; restructured all configs to inherit from it
  • Added JohnsonSU distribution and individual preprocessing per column
  • Adapted embedding layer for new preprocessing pipeline
  • Added unit tests for PRs
  • Fixed column name handling (int → string) in datamodule

v1.1.0 (2025-01-03)

Features

  • Added BaseConfig class to public init
  • Added JohnsonSU distribution support
  • Removed dependency on rotary embeddings

v1.0.0 (2024-12-04)

Initial stable release.