Algorithm performance radar · farther from the centre means better performance. Choose methods to compare. Hover or focus a point for its measured value.
Hover or focus a chart point to inspect its value.
The chart scales each axis from the worst to the best result across all algorithms in this dataset. Changing the selection does not rescale it. Zero is the worst observed result, not absence of capability. Runtime uses reversed log scaling, so faster methods extend farther. Tied axes sit at 0.5.
Single-seed benchmark snapshots. Shapes and polygon areas are not overall scores. Axes are normalized independently for each dataset; compare raw metrics across datasets. Transductive accuracy evaluates a classifier on coordinates fitted to all samples, not a held-out embedding.