Issue: Feature-level p-values are not generated by the Spot-the-diff algorithm

[Gaurav-Sahu-TA]

Description:
Feature-level p-values are not generated by the Spot-the-diff algorithm during drift detection, despite proper implementation.

No specific parameters or options were found to control the generation of feature-level p-values

[arnaudvl]

Hi @Gaurav-Sahu-TA . The Spot-the-diff detector doesn't produce feature-level p-values as it's a multivariate detector over all features. It does however learn the difference between the distributions of the reference and test sets via the kernel coefficients. See e.g. the MNIST and Wine example: https://docs.seldon.io/projects/alibi-detect/en/stable/examples/cd_spot_the_diff_mnist_wine.html.

[Gaurav-Sahu-TA]

Hi @arnaudvl , Are there any existing drift detection techniques specifically designed to identify changes at the feature level in time series data?