Table 1.
Comparing GTED with Well-Known Graph Kernals
| Kernel/data set | Mutag | Enzymes | NCI1 | PTC |
|---|---|---|---|---|
| WL | 84.50 (±2.16) | 53.75 (±1.37) | 84.76 (±0.32) | 59.97 (±1.60) |
| WL-Edge | 82.94 (±2.33) | 52.00 (±0.72) | 84.65 (±0.25) | 60.18 (±2.19) |
| SP | 85.50 (±2.50) | 42.31 (±1.37) | 73.61 (±0.36) | 59.53 (±1.71) |
| Graphlet | 82.44 (±1.29) | 30.95 (±0.73) | 62.40 (±0.27) | 55.88 (±0.31) |
| p-RW | 80.33 (±1.35) | 28.17 (±0.76) | TIMED OUT | 59.85 (±0.95) |
| MLG | 84.21 (±2.61) | 57.92 (±5.39) | 80.83 (±1.29) | 63.62 (±4.69) |
| GTED | 90.12 (±4.48) | 59.66 (±1.84) | 65.83 (±1.14) | 59.08 (±2.11) |
GTED, graph traversal edit distance; MLG, Multiscale Laplacian Graph kernel; p-RW, p-Random Walk; SP, shortest path; WL, Weisfeiler–Lehman. The best accuracy on each dataset is shown in bold.