fix CFRL tabular example

[RobertSamoilescu]

This PR fixes an error in the AE training procedure.
Previously, the AE target was set as follows:

trainset_input = heae_preprocessor(X_train).astype(np.float32)
trainset_outputs = {
    "output_1": X_train_ohe[:, :len(numerical_ids)]
}
...

where

preprocessor = ColumnTransformer(
    transformers=[
        ("cat", cat_transf, categorical_ids),
        ("num", num_transf, numerical_ids),
    ],
    sparse_threshold=0
)
X_train_ohe = preprocessor.transform(X_train)

This wrongly assumed that the preprocessor outputs the numerical columns and then the categorical ones.

To fix the issue, one can use the output of the heae_preprocess which always outputs first the numerical columns and then the categorical ones.

trainset_input = heae_preprocessor(X_train).astype(np.float32)
trainset_outputs = {
    "output_1": trainset_input[:, :len(numerical_ids)]
}

Might be related to the issue #573

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov]

Codecov Report

Merging #651 (034a910) into master (a3d9023) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #651   +/-   ##
=======================================
  Coverage   80.46%   80.46%           
=======================================
  Files          77       77           
  Lines       10635    10635           
=======================================
  Hits         8557     8557           
  Misses       2078     2078           

Impacted Files	Coverage Δ
alibi/explainers/cfproto.py	75.53% <0.00%> (ø)