v0.7.2

What's Changed

New element embeddings

• CHGNet (chgnet): 64D node features from the charge-informed neural network potential, averaged over MP-20
• MatSciBERT (matscibert): 768D token embeddings from a BERT model pre-trained on materials science literature
• Chemeleon (chemeleon): 512D post-GNN node features at t=0 from the generative crystal structure diffusion model, averaged over MP-20

Dependency change

• Removed pymatgen dependency — Mendeleev numbers are now stored as a local dictionary, eliminating the heavyweight pymatgen requirement (closes #5)

Other

• Added composition distance example script (examples/composition_distance.py)
• Fixed docs deployment race condition on gh-pages
• Excluded example scripts from ruff checks
• Regenerated cosine similarity heatmaps (16 PNGs + 14-frame GIF)

v0.7.1

What's Changed

Bug Fix: MLIP embedding dimensions

• MACE-MP-0: Fixed from 256D to 128D. Was concatenating descriptors from all interaction layers; now correctly extracts only the last layer's scalar invariants via get_descriptors(invariants_only=True, num_layers=1)
• SevenNet: Fixed from 64D to 128D. Was hooking reduce_input_to_hidden (post-projection); now correctly hooks the last equivariant gate output to capture full scalar (l=0) node features
• ORB-v2 (256D) was already correct

Other fixes

• Fixed MACE, SevenNet, and ORB citation author names and paper URLs
• SevenNet citation now uses open-access arXiv preprint

Thanks to @lizhenzhupearl for flagging the dimension issues.

v0.7.0

What's Changed

New: MLIP Element Embeddings

Element-level embeddings from three Machine Learning Interatomic Potentials, derived by averaging atom-level representations over MP-20 structures:

• MACE-MP-0 (mace_mp0): 256-dimensional invariant descriptors from the final equivariant message-passing layer
• SevenNet (sevennet): 64-dimensional node features from 7net-0 at the energy readout input
• ORB-v2 (orb_v2): 256-dimensional node features from the energy prediction head input

Usage: Embedding.load_data("mace_mp0")

New: Visualisation Examples

• Cosine similarity heatmaps for all 13 multi-dimensional embeddings (examples/2d_cosine/)
• Animated GIF cycling through all embedding schemes
• 3D element embedding maps (UMAP, PCA, t-SNE) for 8 embedding schemes (examples/3d_maps/)
• Composite visualisation: high-dimensional embedding vectors alongside 3D reduced map

Maintenance

• Renamed Publication/ to publications/
• Added uv to installation instructions
• Fixed dead documentation link in README
• Added extraction scripts for reproducibility (scripts/)

v0.6.2

What's Changed

Maintenance & CI

• Update all pre-commit hooks (ruff v0.5.5→v0.15.2, markdownlint, codespell, pyright, etc.)
• Fix string concatenation bugs with missing spaces in error messages
• Fix ruff lint/format violations across source and notebooks
• Remove Codecov integration (badge + CI upload step)
• Replace black badge with ruff badge
• Remove stale dev dependencies superseded by ruff (black, isort, flake8, autopep8, pyupgrade, nbqa)
• Remove unused typing-extensions runtime dependency
• Fix ruff exclude path and move to top-level [tool.ruff] config
• Fix semantic_release config to reference pyproject.toml
• Add push trigger to CI workflow so badge reflects main branch status
• Fix uv pip show flag compatibility with newer uv
• Fix Windows CI Tcl/Tk errors by setting MPLBACKEND=Agg
• Bump pypa/gh-action-pypi-publish to 1.12.4

v0.6.1

What's Changed

🤷‍♂️ Other Changes

• Overhaul pre-commit by @AntObi in #166
• Fix parse_species to handle non-integer oxidation states by @AntObi in #170

Full Changelog: v0.6...v0.6.1

v0.6

What's Changed

🛠 Enhancements

• Documentation updates and Species updates by @AntObi in #161

🤷‍♂️ Other Changes

• Preparing a new release by @AntObi in #163
• Update latest_docs.yml by @AntObi in #164

Full Changelog: v0.5...v0.6

v0.5

What's Changed

• XenonPy and CGNF embeddings are now available in ElementEmbeddings
• SpeciesEmbedddings are now supported with functions updated to support

🤷‍♂️ Other Changes

• Species by @AntObi in #100
• Xenonpy embeddings by @AntObi in #141
• Update ci.yml by @AntObi in #143
• Add cgnf embeddings by @aronwalsh in #145
• Add literature references to the README.md by @AntObi in #146
• Plotter updates by @AntObi in #153
• Update version number by @AntObi in #154

New Contributors

• @aronwalsh made their first contribution in #145

Full Changelog: v0.4.1...v0.5

v0.4.1

Package requirements have been loosened for better compatibility in python environments.

What's Changed

🤷‍♂️ Other Changes

• Loosen requirements by @AntObi in #120

Full Changelog: v0.4...v0.4.1

v0.4

What's Changed

• Publication folder now updated with SMACT-crystal structure prediction scripts
• oliynyk.json has been replaced with oliynyk.csv which contains the feature labels
• oliynyk_preprocessed.csv has been added with the missing values imputed which is now the default oliynyk embedding

Automated release notes

• Publication updates by @AntObi in #87

Full Changelog: v0.3.1...v0.4

v0.3.1

What's Changed

• Removed warning about standardisation when the Embedding class is instantiated

• composition_featuriser allows custom columns to be used for featurising formulas

• Remove warning by @AntObi in #85

• Comp updates by @AntObi in #86

Full Changelog: v0.3.0...v0.3.1