Table 1 In-distribution performance for the formation energy prediction

From: Probing out-of-distribution generalization in machine learning for materials

Metric

Dataset

ALIGNN

GMP

LLM-Prop

XGB

RF

MAE (eV/at)

MP

0.033

0.052

0.063

0.078

0.090

 

JARVIS

0.036

0.081

0.068

0.074

0.099

 

OQMD

0.020

0.038

0.045

0.070

0.065

R2

MP

0.996

0.992

0.981

0.979

0.970

 

JARVIS

0.995

0.985

0.982

0.981

0.968

 

OQMD

0.998

0.995

0.995

0.987

0.985

  1. Models are arranged in the ascending order of MAEs of the MP dataset from left to right. Best performance is highlighted in bold. The MP, JARVIS, and OQMD datasets contain 146k, 76k, and 1M entries, respectively.
Back to article page