Table 16 stratified brier scores and log loss values-breast cancer dataset.

From: Distilling knowledge from graph neural networks trained on cell graphs to non-neural student models

Method/Data

Stratified Brier Score

Log Loss

Before Calibration

Class 0: 0.01773

Class 1: 0.015629

Class 2: 0.011969

Overall: 0.015111

Class 0: 0.03479

Class 1: 0.05215

Class 2: 0.02904

Overall:0.038663

Isotonic Regression

Class 0: 0.016961

Class 1: 0.0156

Class 2: 0.012267

Overall: 0.0149

Class 0: 0.037611

Class 1: 0.055318

Class 2: 0.030436

Overall: 0.041

Temp Scaling - Reduces Brier Score

Class 0: 0.01753

Class 1: 0.01581

Class 2: 0.01187

Overall: 0.0150708

Class 0: 0.03534

Class 1: 0.05317

Class 2: 0.02959

Overall: 0.039368

Temp Scaling - Reduces Log Loss

Class 0: 0.01778

Class 1: 0.01562

Class 2: 0.01199

Overall: 0.015127

Class 0: 0.03474

Class 1: 0.05206

Class 2: 0.02899

Overall: 0.03860

Back to article page