modl-uclouvain / modnet-matbench

Data repository accompanying De Breuck et al., "Robust model benchmarking and bias-imbalance in data-driven materials science: a case study on MODNet"
https://doi.org/10.1088/1361-648X/ac1280
MIT License
2 stars 2 forks source link

Use the correct featurizer preset for composition-only tasks #19

Closed ml-evs closed 2 years ago

ml-evs commented 2 years ago

At some point we had to specify that the DeBreuck2020Featurizer should use fast_oxid=True for these benchmarks, which meant that we had to manually create the featurizer in run_benchmark.py rather than letting MODNet figure out which one to use. This meant that the few composition-only tasks could not be featurized directly with this script.

This PR also: