Top-K Accuracy
Metric computing the per-head top-k accuracy for
classification: the fraction of samples whose true class falls within the k
highest-scoring predictions. With k = 1 this is ordinary top-1 accuracy.
Configuration
Selected as name: top-k-acc.
Option |
Description |
Type |
Default |
|---|---|---|---|
|
Number of top-scoring classes considered a correct prediction. |
int |
|
services:
metrics:
select:
- name: top-k-acc
kwargs: { k: 3 }