Prediction explanations: multiclass catboost support

[dsherry]

Fixes #2136

The following code works after this PR:

import evalml
X, y = evalml.demos.load_wine()
assert len(y.to_series().unique()) == 3
pipeline = evalml.pipelines.MulticlassClassificationPipeline(['Imputer', 'CatBoost Classifier'])
pipeline.fit(X, y)
print(evalml.model_understanding.prediction_explanations.explain_predictions(
    pipeline=pipeline, input_features=X, y=y, indices_to_explain=[0]))

which outputs

CatBoost Classifier w/ Imputer

{'Imputer': {'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'CatBoost Classifier': {'n_estimators': 10, 'eta': 0.03, 'max_depth': 6, 'bootstrap_type': None, 'silent': True, 'allow_writing_files': False}}

	1 of 1

		Class: class_0
		
		 Feature Name     Feature Value   Contribution to Prediction
		============================================================
		    proline          1065.00                  ++            
		    alcohol           14.23                   +             
		color_intensity       5.64                    +             
		
		
		Class: class_1
		
		 Feature Name     Feature Value   Contribution to Prediction
		============================================================
		    alcohol           14.23                   --            
		    proline          1065.00                  --            
		color_intensity       5.64                    --            
		
		
		Class: class_2
		
		Feature Name    Feature Value   Contribution to Prediction
		==========================================================
		   proline         1065.00                  -             
		 flavanoids         3.06                    -             
		total_phenols       2.80                    -             

[codecov]

Codecov Report

Merging #2224 (3c35a5e) into main (bc8460a) will decrease coverage by 0.1%.
The diff coverage is 100.0%.

@@            Coverage Diff            @@
##             main    #2224     +/-   ##
=========================================
- Coverage   100.0%   100.0%   -0.0%     
=========================================
  Files         288      288             
  Lines       24495    24489      -6     
=========================================
- Hits        24477    24470      -7     
- Misses         18       19      +1     

Impacted Files	Coverage Δ
...s/prediction_explanations_tests/test_algorithms.py	98.3% <ø> (-0.9%)	⬇️
...derstanding/prediction_explanations/_algorithms.py	98.9% <100.0%> (-<0.1%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bc8460a...3c35a5e. Read the comment docs.

[dsherry]

After deleting this check, this test does verify that multiclass catboost works. I verified this by changing the multiclass catboost impl to error, and saw that this test errored out.

[dsherry]

I didn't see this message being used anywhere. I think its left over from a previous change.

[dsherry]

I don't think this is true anymore with shap >= 0.36.0!

[dsherry]

I think this was the wrong issue. This is a discussion about xgboost.

[chukarsten]

Looks good!

[freddyaboulton]

Looks good @dsherry ! Can you update the model understanding docs? There's a sentence saying we don't support prediction explanations for catboost multiclass.

[bchen1116]

Nice!