Comparison with pymatgen, matminer, automatminer, matbench

capoe / benchml

ML benchmarking and pipeling framework

Apache License 2.0

15 stars 6 forks source link

Comparison with pymatgen, matminer, automatminer, matbench #18

Open sgbaird opened 2 years ago

sgbaird commented 2 years ago

Would be nice to see a short description in the README describing what additional benefits this repo offers relative to pymatgen, matminer (featurization), automatminer (hyperparameter optimization/pipelines), and matbench (materials benchmarking) (non-affiliated for all of these). What are the differences/advantages?

capoe commented 2 years ago

Quick answer in place: BenchML focuses on

... flexible pipelining with complex dependency graphs (including, for example, "decorator" transforms to build ensembles or conformal predictors),
... not just molecules and materials (see for instance examples/demo),
... specialized statistical techniques not commonly found elsewhere, such as feature network and random-matrix filters.