Add a translation layer between TaijaData.jl and CounterfactualExplanations.jl

[kmariuszk]

CounterfactualExplanations

• Add a translation layer which transfers (X, y) tuple into CounterfactualData object (maybe use extensions?)
• Test it
• Document it

Problems:

• How to not lose additional information (e.g., CounterfactualExplanations.CounterfactualData(X, y; domain=(-1.0, 1.0) domain information)?
  • Either add recalculate it in the translation layer or return more data (i.e., X, y, domain)

[kmariuszk]

@pat-alt I noticed there was only few functions that actually pre-process the data additionally to loading it (e.g., specifying the domain of the dataset or specifying the standardization - these are the parts I left commented-out in the #9) and I question if we actually need this extension layer. Do you think it's necessary?

[pat-alt]

@kmariuszk coming back to this now and realizing I missed this comment. I still think we need a translation layer, but unsure whether to place it here or in CE.jl. Let's discuss during our next call.

[pat-alt]

So from what I can tell, a translation layer would be needed to ensure that:

• credit_default data has adequately encoded categorical features
• cifar_10, fashion_mnist and mnist get domain constraints