Table 5 Final regression and classification model predictions of the experimental holdout set.

From: A database of experimentally measured lithium solid electrolyte conductivities evaluated with machine learning

Composition

Measured conductivity (log10(σ))

CrabNet regression prediction (log10(σ))

CrabNet classifier prediction (log10(σ) ≥ 4)

Li10.35Ge1.35P1.65S1250

−1.85

−3.60

1

Li10.35[Sn0.27Si1.08]P1.65S1251

−1.96

−3.50

1

Li10GeP2S11.7O0.352

−1.99

−3.06

1

Li10GeP2S11.4O0.652

−2.07

−3.07

1

Li10[Si0.3Sn0.7]P2S1251

−2.09

−2.66

1

Li9.42Si1.02P2.1S9.96O2.0453

−3.49

−3.67

1

Li3.35P0.93S3.5O0.548

−4.04

−2.67

1

Li3.3SnS3.3Cl0.754

−4.49

−3.62

0

Li4.3AlS3.3Cl0.749

−5.09

−7.14

0

Li3P5O1455

−6.04

−7.73

0

LiAlP2O756

(Very low)

−6.32

0

  1. CrabNets with transfer learning are trained on all 403 unique compositions and the associated log10(σ) or classification target at room temperature. The experimentally measured log10(σ) of each of the 11 materials in the holdout set are given alongside a predicted log10(σ) and conductivity class for the material from the final models, the boundary against which the classification is performed has been marked in black.
Back to article page