matbench-discovery

📦 Package fix release. Drops unnecessary numpy>2 pin.

What's Changed

🐛 Bug Fixes

• Fix model detail pages for models with array of training sets and non-string notes in YAML metadata by @janosh in https://github.com/janosh/matbench-discovery/pull/138

🌐 Site Upgrades

• Show single-model energy parity plots on model detail pages by @janosh in https://github.com/janosh/matbench-discovery/pull/136
• Move element-projected convex hull distance errors to model detail pages by @janosh in https://github.com/janosh/matbench-discovery/pull/137

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v1.3.0...v1.3.1

What's Changed

🆕 New Model

• Test mace_mp_0b by @janosh in https://github.com/janosh/matbench-discovery/pull/119
• Orb models by @DeNeutoy in https://github.com/janosh/matbench-discovery/pull/133

🛠 Enhancements

• data.py add ase_atoms_(to|from)_zip by @janosh in https://github.com/janosh/matbench-discovery/pull/120
• Allow multiple training sets per model in YAML metadata by @CompRhys in https://github.com/janosh/matbench-discovery/pull/132
• Add metrics table download buttons by @janosh in https://github.com/janosh/matbench-discovery/pull/135

🐛 Bug Fixes

• Type hint optimizer and filter in test_mace and test_7net correctly, fix pickled PatchedPhaseDiagram and compile_wbm_test_set.py by @CompRhys in https://github.com/janosh/matbench-discovery/pull/122
• Working save trajectory code with energies also being saved for test mace. by @CompRhys in https://github.com/janosh/matbench-discovery/pull/123
• Correct batch_size of SevenNet-0 by @YutackPark in https://github.com/janosh/matbench-discovery/pull/129
• Restore MPTrj EDA to working order by @CompRhys in https://github.com/janosh/matbench-discovery/pull/130

📖 Documentation

• Add n_estimators to model_metadata.yml by @janosh in https://github.com/janosh/matbench-discovery/pull/115
• Update table caption for landing page by @CompRhys in https://github.com/janosh/matbench-discovery/pull/116
• Add individual model detail pages by @janosh in https://github.com/janosh/matbench-discovery/pull/126

🧹 House-Keeping

• Update the function names for wren utils given breaking changes in Aviary by @CompRhys in https://github.com/janosh/matbench-discovery/pull/117

🚧 CI

• Test more plotting scripts in CI by @janosh in https://github.com/janosh/matbench-discovery/pull/127

💡 Refactoring

• Refactor Files, DataFiles and PredFiles to be StrEnums by @janosh in https://github.com/janosh/matbench-discovery/pull/114

New Contributors

• @DeNeutoy made their first contribution in https://github.com/janosh/matbench-discovery/pull/133

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v1.2.0...v1.3.0

What's Changed

• fix: changes to wrenformer test script by @CompRhys in https://github.com/janosh/matbench-discovery/pull/105
• Fix figs/e-form-parity-models showing convex hull distance instead of formation energy R² in subplot titles by @janosh in https://github.com/janosh/matbench-discovery/pull/106
• fix: CHGNet MPTrj training set now published and widely used. by @CompRhys in https://github.com/janosh/matbench-discovery/pull/107
• Add MatterSim predictions by @yanghan-microsoft in https://github.com/janosh/matbench-discovery/pull/104
• Migrate Key enum to use pymatviz.enums.Key by @janosh in https://github.com/janosh/matbench-discovery/pull/110
• Add new /models/mace page with formation energy and raw energy parity plots by @janosh in https://github.com/janosh/matbench-discovery/pull/111
• Update MACE readme by @CompRhys in https://github.com/janosh/matbench-discovery/pull/113
• add SevenNet to Matbench Discovery leaderboard by @YutackPark in https://github.com/janosh/matbench-discovery/pull/112

New Contributors

• @yanghan-microsoft made their first contribution in https://github.com/janosh/matbench-discovery/pull/104
• @YutackPark made their first contribution in https://github.com/janosh/matbench-discovery/pull/112

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v1.1.2...v1.2.0

What's Changed

• Fix test scripts raising wandb.Table AssertionError: columns argument expects list of strings or ints by @janosh in https://github.com/janosh/matbench-discovery/pull/82
• Add GNoME model by @janosh in https://github.com/janosh/matbench-discovery/pull/84
• Model metadata schema validation by @janosh in https://github.com/janosh/matbench-discovery/pull/85
• Extend model metadata schema by @janosh in https://github.com/janosh/matbench-discovery/pull/86
• Pass preds rather than err to v-plots by @CompRhys in https://github.com/janosh/matbench-discovery/pull/87
• Rough solution to recreating plots for unique proto set by @CompRhys in https://github.com/janosh/matbench-discovery/pull/88
• Fill 3% missing MACE preds by @janosh in https://github.com/janosh/matbench-discovery/pull/90
• Fix landing page vs model page metrics mismatch by @janosh in https://github.com/janosh/matbench-discovery/pull/92
• Add pyright pre-commit hook and fix possibly unbound variables by @janosh in https://github.com/janosh/matbench-discovery/pull/93

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v1.1.1...v1.1.2

Hot Fix Release

• Fix failing pkg_is_editable check in PyPI release by @janosh in https://github.com/janosh/matbench-discovery/pull/78

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v1.1.0...v1.1.1

What's Changed

• Update MACE results by @janosh in https://github.com/janosh/matbench-discovery/pull/54
• Add train_mace.py for easier reproducibility by @chiang-yuan in https://github.com/janosh/matbench-discovery/pull/52
• Re-test CHGNet v0.3.0 by @janosh in https://github.com/janosh/matbench-discovery/pull/58
• Update figures to latest MACE checkpoint by @janosh in https://github.com/janosh/matbench-discovery/pull/59
• Make distinction between training set n_structures and n_materials for models trained on DFT trajectories by @janosh in https://github.com/janosh/matbench-discovery/pull/60
• Fix missing stress unit conversion (kBar->eV/A^3) in json_to_extxyz.py by @chiang-yuan in https://github.com/janosh/matbench-discovery/pull/61
• add task_id, calc_id, ionic_step metadata to atoms.info by @chiang-yuan in https://github.com/janosh/matbench-discovery/pull/63
• More extensive MP/WBM data analysis for site and SI by @janosh in https://github.com/janosh/matbench-discovery/pull/64
• UMAP visualization of MP and WBM data by @pbenner in https://github.com/janosh/matbench-discovery/pull/65
• Add flags to CHGNet and MACE test scripts to run in static mode (no relaxation) by @janosh in https://github.com/janosh/matbench-discovery/pull/70
• Rerun CHGNet with new ASE FrechetCellFilter (prev ExpCellFilter) by @janosh in https://github.com/janosh/matbench-discovery/pull/71
• Rerun M3GNet with new ASE FrechetCellFilter (prev ExpCellFilter) by @janosh in https://github.com/janosh/matbench-discovery/pull/72
• Define StrEnums Key and Task for consistent + easily updatable dataframe columns by @janosh in https://github.com/janosh/matbench-discovery/pull/74
• Add metrics analysis excluding WBM materials with duplicate/MP-matching structure prototype by @janosh in https://github.com/janosh/matbench-discovery/pull/75

New Contributors

• @chiang-yuan made their first contribution in https://github.com/janosh/matbench-discovery/pull/52

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v1.0.0...v1.1.0

What's Changed

• Add MACE by @janosh in https://github.com/janosh/matbench-discovery/pull/48
• Add ALIGNN FF (aborted) by @janosh in https://github.com/janosh/matbench-discovery/pull/47
• Add training set URLs in ModelCard and show training set size in metrics tables by @janosh in https://github.com/janosh/matbench-discovery/pull/49
• Add MEGNet RS2RE to UIP->MEGNet comparison by @janosh in https://github.com/janosh/matbench-discovery/pull/50
• Update site to match arXiv by @janosh in https://github.com/janosh/matbench-discovery/pull/51

Big thanks to @CompRhys, @pbenner and @chiang-yuan for their help getting to v1. 🙏

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v0.1.4...v1.0.0

What's Changed

• Update site to show ALIGNN results #38
• ALIGNN test on WBM data #37
• Add PDF versions of all SI figures #36
• Display largest-error-scatter-select.svelte in /si #35
• Fix KeyError('wyckoff_spglib') in fetch_process_wbm_dataset.py #34
• Fix compute_struct_fingerprints.py: cannot insert material_id, already exists #31
• Fix fetch_process_wbm_dataset.py unnecessarily loading WBM initial structures from disk #29
• Fix loading 2023-02-07-ppd-mp.pkl.gz #26
• Fix downloading compressed Figshare data #21
• Run scripts in CI #19
• Prompt to download missing data files #15
• Add upload_to_figshare.py and replace local data-loading with cached Figshare downloads #13
• DRY workflows #8
• Add Files, DataFiles and PredFiles classes in matbench_discovery/data.py #7
• fix path issues and using JSON pandas reader for CSV files in train_(cgcnn|wrenformer).py (closes #43) #43
• Fix KeyError('wyckoff_spglib') in fetch_process_wbm_dataset.py (#34) #32 #33
• Fix compute_struct_fingerprints.py: cannot insert material_id, already exists (#31) #30
• Fix fetch_process_wbm_dataset.py unnecessarily loading WBM initial structures from disk (#29) #28
• fix AttributeError: 'float' object has no attribute 'split' in fetch_process_wbm_dataset.py (closes #27) #27
• Fix loading 2023-02-07-ppd-mp.pkl.gz (#26) #23
• Fix downloading compressed Figshare data (#21) #20 #17
• Run scripts in CI (#19) #16
• Prompt to download missing data files (#15) #14 #14
• Add upload_to_figshare.py and replace local data-loading with cached Figshare downloads (#13) #10 #12 #11
• support model_name as array in metadata.yml files to share metadata between multiple types of a model c8eaebd
• include legacy MP energy corrections in data/wbm/2022-10-19-wbm-summary.csv, use them to remove old and apply new corrections in test_megnet.py 8166801
• fix join_chgnet_results.py by removing code to apply MP2020 corrections 01658ad
• add columns wyckoff_spglib and spacegroup_symbol to data/mp/2023-01-10-mp-energies.csv 01e88f0
• add model metrics table when only validating the 10k most stable predictions for each model c4ca186
• run model combos (m3gnet|chgnet)+megnet through test_megnet.py fbe847b
• split analyze_model_failure_cases.py into two scripts, new one is analyze_elements.py 0fad3bd
• add scatter plot of largest errors averaged over models vs DFT hull distance 457abf0
• add scripts/compute_struct_fingerprints.py to generate matminer SiteStats fingerprints for all MP+WBM structures 7946b5e
• split model run times into train and test contribs b8a18d8
• add models/chgnet/{test_chgnet,join_m3gnet_results}.py da39074
• git surgery purging all large CSV files from repo history 4a90dee
• add scripts/compute_projections.py eec1cb4
• make horizontal versions of roc-models-2x4.pdf and model-run-times-bar.pdf 551050e
• add changelog.md 0621282
• add models/chgnet/ctk_struct_traj.py and models/wrenformer/analyze_wrenformer.py c9fed5a
• add data/mp/get_mp_traj.py to save a snapshot of all MP ionic steps on 2023-03-15 to be released as MBD canonical training set 76647e3
• move load_df_wbm_preds+PRED_FILES from matbench_discovery/{data->preds}.py 387184c
• add site/src/figs/(largest-fp-diff-each-error-models|largest-each-errors-fp-diff-models).svelte 4b6e83a
• move /paper/preprint to /paper 8a0c3bb
• add global STABILITY_THRESHOLD to consistently parametrize across the codebase when materials count as thermodynamically stable 70b2b1d
• plot easy vs hard structures (for all models) norm of SiteStats fingerprint difference before/after relaxation a6bfa74
• add missing doc strings 1ae72c0
• fix direct data file download URLs on /contribute page 83736ad
• add git submodule paper to prepare journal submission in separate repo 3ac72e6
• add figs/cumulative-mae-rmse.svelte displayed in /si 8798786
• rename ElemCountInset to PtableInset and add props precision, show_percent, unit abfc0ac
• add scripts/ctk_structure_viewer.py fa1a439
• +layout.svelte add <svelte:head> with <title> and <meta description> 5d38918
• refactor data loading in model test scripts d564ade
• add CmdPalette invoked with cmd+k for keyboard-only site navigation 59010fa
• improve preprint results + discussion wording d27f1c4
• fix outdated numbers in data/wbm/readme.md 79c8042
• change SI plot of largest model errors: Predicted vs. DFT hull distance colored by model disagreement a5b3211
• show figs/element-prevalence-vs-error.svelte on /si page 6771462
• rm frontmatter.yml after consolidating with citation.cff 228be00
• rename data/wbm/(analysis->eda).py 0f2410d
• make wide tables horizontally scrollable on mobile screens fc20ccf
• toggle site tips on cmd+j, update deps, tweak copy (abstract, metrics-table-first-10k, readme) 137c414
• add date_published field to model metadata cddda31
• fix /contribute dfa6c24
• allow toggling MP+WBM element prevalence ptable heat maps between count modes 'occurrence' and 'composition' 1fe5bcb
• add material_id to figs/mp-elemental-ref-energies.svelte hover tooltip c87f0b2
• bump @rollup/plugin-yaml to v4.1 and refactor citation.cff import using new extensions option cc08c78
• tweak contributing guide b3ea8d3
• upload @pbenner's ALIGNN checkpoint to Figshare and link from models/alignn/readme.md da56377
• start preprint fd8355d
• rename pages /about-test-test-set to /about-the-data and /how-to-contribute to /contribute e41ca32

Full Changelog: https://github.com/janosh/matbench-discovery/compare/v0.1.4...v0.1.5