Table 1 A summary of the three winning models of the competition with the test-set root mean square log error (RMSLE) and mean absolute error (MAE) of the formation energy and bandgap energy

From: Crowd-sourcing materials-science challenges with the NOMAD 2018 Kaggle competition

Ranking

ML representation + regression method

Formation energy

Bandgap energy

Room mean square log error

Mean absolute error (meV/cation)

Root mean square log error

Mean absolute error (meV/cation)

First

n-gram + KRR

0.020 (0.021a)

14 (15a)

0.078 (0.077a)

106 (114a)

Second

c/BOP + LGBM

0.022

15

0.081

93

Third

SOAP + NN

0.021

13

0.083

98

  1. aDetermined by using the 4-gram representation instead of an ensemble of 3-gram and 4-gram that was used in the actual competition
Back to article page