Log10 (All unique sites mean squared displacements @ 300K - Ų) - log_msd_all_300

ARFS Top features

ARFS selected descriptors

---

Correlation analysis

Distance correlation

Dependency graphs

Feature learnability

---

Model performance

5-Fold CV Metrics overview

RF - MATMINER

	train_rmse	test_rmse	train_errors	test_errors	train_r2	test_r2
mean	0.03926	0.10432	0.02832	0.07562	0.9841	0.8875
min	0.0388	0.0999	0.0281	0.0748	0.9837	0.8737
max	0.0396	0.1094	0.0286	0.0769	0.9846	0.9006
std	0.00026533	0.00393517	0.000203961	0.000724983	0.000289828	0.00901887

RF - MATMINER+LOBSTER

	train_rmse	test_rmse	train_errors	test_errors	train_r2	test_r2
mean	0.03856	0.10306	0.0278	0.0747	0.98468	0.8902
min	0.038	0.098	0.0275	0.0728	0.9844	0.8758
max	0.0391	0.1089	0.0282	0.077	0.9852	0.9015
std	0.000361109	0.00464052	0.000289828	0.00146697	0.000318748	0.00958374

MODNet - MATMINER

	train_rmse	test_rmse	train_errors	test_errors	train_r2	test_r2
mean	0.0393	0.08646	0.0251	0.05858	0.98392	0.92268
min	0.0358	0.0816	0.0231	0.0572	0.9773	0.9167
max	0.0471	0.0891	0.0305	0.0603	0.9867	0.9336
std	0.00408363	0.00294252	0.00275318	0.00114263	0.00346318	0.00616876

MODNet - MATMINER+LOBSTER

	train_rmse	test_rmse	train_errors	test_errors	train_r2	test_r2
mean	0.04068	0.08644	0.02634	0.0592	0.98262	0.92286
min	0.0342	0.0823	0.0222	0.0557	0.9774	0.9172
max	0.047	0.0901	0.0304	0.0632	0.988	0.9284
std	0.00534318	0.00298101	0.00351659	0.00300267	0.00447946	0.00402423

Corrected resampled t-test on 10-fold CV

Summary

	t_stat	p_value	significance_stars	d_av	rel_improvement	percent_folds_improved
RF	2.44741	0.0184566	*	0.360562	1.48878	80
MODNet	1.77032	0.0552254		0.715069	3.85596	90

---

Model Explainer

PFI

SHAP

---

SISSO Models

Rung 1

1D descriptor

\[\begin{align*} & log\_msd\_all\_300 = -1.423 \\ & -0.187\left(\ln{ G4_{0.005\_1.0\_1.0} }\right) \end{align*}\]

2D descriptor

\[\begin{align*} & log\_msd\_all\_300 = -1.578 \\ & -0.173\left(\ln{ G4_{0.005\_1.0\_1.0} }\right) \\ & + 0.007\left(\frac{ site_{ElectronAffinity} }{ site_{Column} } \right) \end{align*}\]

Rung 2

1D descriptor

\[\begin{align*} & log\_msd\_all\_300 = -1.527 \\ & -5.07\left(\left(\ln{ site_{AtomicWeight} }\right) \left(\frac{ Gaussian_{center\_3.0\_width\_1.0} }{ Average_{bond\_length} } \right)\right) \end{align*}\]

2D descriptor

\[\begin{align*} & log\_msd\_all\_300 = -1.762 \\ & + 0.005\left(\left(\frac{ site_{ElectronAffinity} }{ site_{Column} } \right) + \left(site_{ElectronAffinity} BOOP_{Q\_l\_2}\right)\right) \\ & + 0.191\left(\left(\sqrt{ local_{difference\_in\_Electronegativity} }\right) - \left(\ln{ G4_{0.005\_1.0\_1.0} }\right)\right) \end{align*}\]

---

Misc

ARFS n-iter convergence checks

MAE/ fold from 10-fold CV

Alternative visual summary of input data for t-test