Data repository accompanying De Breuck et al., "Robust model benchmarking and bias-imbalance in data-driven materials science: a case study on MODNet"
2
stars
2
forks
source link
Featurizing from scratch does not work for composition-only tasks #6
Closed
ml-evs closed 3 years ago
There's some broken logic that causes featurization to fail for composition-only tasks:
https://github.com/ml-evs/modnet-matbench/blob/30cccbee06f3405846ec8bbc598da9578d54200c/run_benchmark.py#L71-L79