KNN#

class scio.scores.KNN(*, act_norm=2.0, mode='raw', k=?, index_metric='l2')[source]#

Bases: BaseScoreClassif

KNN for classification.

Nonconformity is the distance to \(k\)-th neighbor in the chosen latent space, after activations normalization. Note that it is independant of the predicted class.

Parameters:

k (int) – Number of nearest neighbors to consider.
index_metric (IndexMetricLike) – Kind of metric to use for nearest neighbors search. See IndexMetric.
mode – See BaseScoreClassif.
act_norm – See BaseScore. Now defaults to 2.0.

Note

In [SMZL22], authors operate solely on the penultimate layer activations. Here, we stack all recorded activations, which is equivalent if and only if the network only records the penultimate layer.

References

[SMZL22]

Yiyou Sun, Yifei Ming, Xiaojin Zhu, and Yixuan Li. Out-of-Distribution detection with deep nearest neighbors. In Proceedings of the 39th International Conference on Machine Learning, volume 162 of Proceedings of Machine Learning Research, 20827–20840. 2022. URL: https://proceedings.mlr.press/v162/sun22d.html.