Table 1 Test accuracies in MAEs for the formation energy (Ef), the bandgap (Eg and nonzero gap \({E}_{{{{\rm{g}}}}}^{{{{\rm{nz}}}}}\)), the bulk modulus (KVRH), the shear modulus (GVRH), and the refractive index (n) from ECSTL/ECMTL models and prior works on the materials project data set.

From: Element-wise representations with ECNet for material property prediction and applications in high-entropy alloys

Property

Ntrain

units

ECSTL

ECMTL

MODNeta

MEGNetb

SISSOa

Ef

60,000

eV/atom

0.076

0.084

0.044

0.028

0.299

Eg

60000

eV

0.164

0.227

0.34

0.33

0.75

\({E}_{{{{\rm{g}}}}}^{{{{\rm{nz}}}}}\)

37179

eV

0.27

0.27

0.45

0.38

0.86

KVRH

4722

\({\log }_{10}({{{\rm{GPa}}}})\)

0.05

0.05

 

0.05

 

GVRH

4722

\({\log }_{10}({{{\rm{GPa}}}})\)

0.049

0.046

 

0.079

 

n

3272

 

0.046

 

0.05

0.08

0.12

  1. aResults for single-property learning from the MODNet and SISSO models in ref. 30.
  2. bPerformance listed in the original MEGNet paper in ref. 31.
  3. The training set size is represented as Ntrain, and the underlined values correspond to the lowest error for each task.
Back to article page