Skip to content

1.8.0 Release

Latest
Latest

Choose a tag to compare

View all tags
@mauvais2 mauvais2 released this 28 Apr 22:27
· 8 commits to master since this release
1.8.0
e77c696
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Highlights

  • Updates AMPL to use py3.10, DeepChem 2.8 and related packages
  • Standardizing Python project setup using uv.
  • Publish the AMPL Python package to PyPI so users can install it directly without building it locally, improving accessibility, and supporting broader adoption through a trusted distribution channel.
  • Changed hyperopt to optuna for hyperparameter optimization.
  • Different calculations for model performance metrics. Metric calculations are now performed differently, previous results will not be reproduced in some circumstances.
  • Enhanced transformer workflows.
  • Expanded feature generation capabilities.
  • Improved transfer learning support across the project.

Enhancements

  • Expanded transfer learning support by enabling pretrained AMPL models to serve as feature encoders. (#424)
  • Added scalable RDKit and Mordred feature sets, plus support for fitting and applying transforms on larger or unlabeled datasets. (#424)
  • Removed the deprecated UMAP feature transformer. (#424)
  • Migrated hyperparameter tuning from Hyperopt to Optuna. (#431)
  • Refactored search logic, checkpointing, and related configs. (#431)
  • Updated dependencies, tests, and documentation. (#431)
  • Standardized Python setup with uv for reproducible, efficient development. (#429)
  • Added multi-platform support in pyproject.toml for CPU, CUDA, ROCm, and MPS/mchip. (#429)
  • Updated CI, code, and tests to improve reliability. (#429)
  • Automated package validation and publishing to TestPyPI and PyPI. (#429)
  • Enhanced docs and added platform-specific Dockerfiles for easier deployment. (#429)
  • Handle NaN/Inf and partially update testing suite to be more robust (#432)
    • Strengthens handling of invalid and non-finite predictions across evaluation, model comparison, and hyperparameter tuning, with configurable filtering
    • Improved reporting, and safer model selection. It also improves test reliability and local portability through expanded coverage, cleanup fixes, and macOS-specific compatibility updates
  • Fix remove_outlier_replicates to exclude rows with NaN values in the response column before outlier removal. (#420)
  • Remove molvs dependency (#428)
    • Replaced molvs-based SMILES standardization with RDKit’s built-in functionality, simplifying structure processing and aligning with a maintained toolkit.
    • Removed the unmaintained molvs dependency from AMPL environments, reducing maintenance overhead and improving environment stability.
  • Change to the BSEP graphconv model to improve the test speed dramatic… . (#426)

Bug Fixes

  • Updated remove_outlier_replicates documentation to clarify that it removes rows with missing response values before performing outlier detection. (#421)
  • Fix tests in GitHub CI (#411)

Maintenance

  • libpng removed from Ubuntu mirrors (#423)
    • Added apt-get update before installing libpng to refresh package indexes and restore install reliability.
    • Mitigates recent mirror or security-update related package resolution failures during setup.
  • Merged coverage reports from all test jobs and uploaded a single combined report to Codecov, ensuring complete unit and integration test coverage.
Assets 2
Loading