Table 3 Understanding how Fitzpatrick17k-C classification performance varies with change in hyperparameters.
From: Investigating the Quality of DermaMNIST and Fitzpatrick17k Dermatological Image Datasets
Hyperparameters optimized on → | Verified | Random (Stratified) | Source A (Atl. Derm.) | Source B (DermaAmin) | FST 3–6 | FST 1–2 & 5–6 | FST 1–4 |
|---|---|---|---|---|---|---|---|
Best Hyperparameters (n_epochs, optim, lr) | (200, Adam, 1e-3) | (200, Adam, 1e-4) | (100, Adam, 1e-4) | (200, SGD, 1e-2) | (100, Adam, 1e-4) | (200, Adam, 1e-4) | (200, Adam, 1e-4) |
Holdout set performance (overall test accuracy) | |||||||
Verified | 4.65% ± 0.00% | 4.34% ± 0.22% | 4.50% ± 0.22% | 3.88% ± 0.96% | 4.50% ± 0.22% | 4.34% ± 0.22% | 4.34% ± 0.22% |
Random Holdout | 20.20% ± 0.32% | 24.05% ± 0.34% | 23.23% ± 0.43% | 19.25% ± 0.57% | 23.23% ± 0.43% | 24.05% ± 0.34% | 24.05% ± 0.34% |
Source A | 14.96% ± 0.77% | 16.38% ± 0.77% | 16.62% ± 0.86% | 14.28% ± 0.30% | 16.62% ± 0.86% | 16.38% ± 0.77% | 16.38% ± 0.77% |
Source B | 4.78% ± 0.20% | 4.50% ± 0.32% | 4.50% ± 0.32% | 5.16% ± 0.20% | 4.50% ± 0.32% | 4.50% ± 0.32% | 4.50% ± 0.32% |
FST 3–6 | 12.65% ± 0.23% | 14.36% ± 0.15% | 14.42% ± 0.13% | 11.80% ± 0.69% | 14.42% ± 0.13% | 14.36% ± 0.15% | 14.36% ± 0.15% |
FST 1–2 & 5–6 | 14.56% ± 0.53% | 17.27% ± 0.38% | 16.99% ± 0.33% | 13.78% ± 0.41% | 16.99% ± 0.33% | 17.27% ± 0.38% | 17.27% ± 0.38% |
FST 1–4 | 11.34% ± 0.66% | 12.07% ± 0.28% | 11.92% ± 0.38% | 8.08% ± 4.58% | 11.92% ± 0.38% | 12.07% ± 0.28% | 12.07% ± 0.28% |
