Queue
Loading…
–
Loading…
How often the model's chosen answer is right, against how confident it was.
Raw model probabilities, with no recalibration. ECE averages the gap between accuracy and mean confidence over 15 equal-width bins, weighted by bin size, as in the evaluation code. Brier is the multi-class form (0 is best, 2 is worst).
| Bin | Items | Accuracy | Mean conf. | Gap |
|---|
Same items, same cutoff. Select a row to switch model.
| Model | Accuracy | ECE | Auto-accept at ≤1% | At current cutoff |
|---|
| Group | Items | Accuracy | Mean conf. |
|---|