Ola Zytek

The modeling functionality will now use `pyreal` explainers to generate the feature contribution explanations. This allows the external library to handle transformation logic

feature_scatter_plot should default to a continuous legend if predictions contains floats, but it does not correctly check for np.floats

We should have a single function to prepare all realapp explainers

Some transformers, like FeatureSelect, will break if not fit without raising a useful error. This should be fixed.

The explainers of a single RealApp object all need to transform the dataset separately, even though this is often the same process between explainers. We should investigate ways to share...

#511 introduced `faiss-cpu` as a dependency to speed up the computation of similar examples. This package may not be as maintained as we'd like, so we should make it an...

Runtime testing determined that when GlobalFeatureImportance explainers are given a training set on init and fit using `fit_on_init`, they take about twice the time to fit and produce an explanation...

For this issue we will add a LIME algorithm option for local feature contributions This involves the following steps: 1. Add the `lime` package to the poetry project (poetry add...

Implement counterfactual explanations

1

Counterfactual explanations are a popular and very useful way of explaining an ML model prediction. To start, we should implement a basic optimization approach, with the potential to expand into...

It seems that unit tests are not thoroughly testing that the values in returned explanations correctly correspond to the explanations. This functionality should be added.