Sampling n random samples per class

[SherlockMones]

To simplify exemplary sampling, it would be awesome to allow users to sample a defined set of images from each existing class.

Example

Consider a multiclass problem with three classes: A, B, C, where an image can only ever belong to one class.
If I am done with training and prediction, I'd like to sample random images per class and apply XAI. Right now, I have to manually sample a random image per class.

Possible solution

Either extend the XAI functionality and offer to automatically sample a random image per class or extend the sampling module by adding a function to sample a number of images per class, and not just a percentage split.