v0.5.0

Major Changes

• Unified High-Level API (CrystALLSearch): Polished the framework into more intuitive API managing the full pipeline, from prototype querying and atomic substitutions to featurization and candidate ranking
• Modular "Community" ML Calculator Adapters: Introduced plug-and-play property predictors with native support for ALIGNN via our MPDD-ALIGNN fork (multi-objective models incl. formation energy & bandgap models) and CHGNet (static evaluation and full structure relaxation)
• Systematized the Feature-Space Analysis: Integrated t-SNE embeddings with automated KMeans (silhouette-optimized) and DBSCAN clustering to separate distinct structural families and extract diverse top candidates. In earlier versions this has been done in Wolfram and R, while v0.5.0 moved everything into Python.
• Built-in 2D & 3D Visualizations: Added interactive plotting tools (plot_2D, plot_3D) for candidate cluster analysis and energy landscape sorting. The data output to custom tools is still as smooth as ever though!
• Tutorials & Pre-Built Cloud Environment: Released comprehensive hands-on Jupyter tutorials (BaSO4 walk-through and hydride discovery with embedded 3D structure models) alongside one-click GitHub Codespaces with everything pre-built.
• Packaging & CI/CD: Modernized setup with pyproject.toml, complete Sphinx documentation to be later pushed to Read The Docs, and automated multi-platform CI testing across Linux, macOS, and Windows.
• See V0.5.0 PR by @ricardonpa in #10 for discussion and design choices made

New Contributors

• @ricardonpa made their first contribution in #1

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