Add StandardScaler to ElasticNet pipelines. #1065
Conversation
Codecov Report
@@ Coverage Diff @@
## main #1065 +/- ##
=======================================
Coverage 99.91% 99.91%
=======================================
Files 188 188
Lines 10286 10296 +10
=======================================
+ Hits 10277 10287 +10
Misses 9 9
Continue to review full report at Codecov.
|
freddyaboulton
changed the title
freddyaboulton
requested review from
angela97lin,
dsherry,
jeremyliweishih and
eccabay
| @@ -56,7 +55,7 @@ def _get_preprocessing_components(X, y, problem_type, estimator_class): | |||
| if (add_datetime_featurizer or len(categorical_cols.columns) > 0) and estimator_class not in {CatBoostClassifier, CatBoostRegressor}: | |||
| pp_components.append(OneHotEncoder) | |||
|
|
|||
| if estimator_class in {LinearRegressor, LogisticRegressionClassifier}: | |||
| if estimator_class.model_family == ModelFamily.LINEAR_MODEL: | |||
There was a problem hiding this comment.
Nice! Could we add a test (testing test_make_pipeline) to make sure :d
There was a problem hiding this comment.
Great! Thanks for thinking of this. As mentioned in slack, I think all non-tree-based models should apply scaling, for now.
@freddyaboulton it would be great to see perf test results on this, perhaps 5 trials on datasets_small_0.yaml with model_family limited to linear. But since this is a small change, not required--if it did introduce a regression we'd catch it before release.
| @@ -114,6 +116,11 @@ def test_make_pipeline(): | |||
| assert isinstance(binary_pipeline, type(BinaryClassificationPipeline)) | |||
| assert binary_pipeline.component_graph == [DropNullColumns, Imputer, DateTimeFeaturizer, OneHotEncoder, StandardScaler, LogisticRegressionClassifier] | |||
|
|
|||
| en_binary_pipeline = make_pipeline(X, y, ElasticNetClassifier, ProblemTypes.BINARY) | |||
There was a problem hiding this comment.
Nice, thanks for this!
|
@angela97lin @dsherry @eccabay I was able to run the performance tests for this change. The results are here. In short, introducing this change increases the number of times EN is picked as the best pipeline without introducing any regressions (fit time or average best pipeline score) so I think we're ok to merge this. |
Pull Request Description
Adds StandardScaler to all ElasticNet pipelines.
After creating the pull request: in order to pass the release_notes_updated check you will need to update the "Future Release" section of
docs/source/release_notes.rstto include this pull request by adding :pr:123.