Pass optional column names to parallel tree function in using the self.feature_names_in attribute. #26208
Conversation
…list if known to the attributes of the tree
There was a problem hiding this comment.
Thank you for the PR!
Based on the discussion in #26140, we do not have an agreed solution yet.
sklearn/ensemble/_forest.py
Outdated
|
|
||
| feature_names_in_ = None | ||
| if isinstance(X, pd.DataFrame): | ||
| feature_names_in_ = X.columns.to_list() |
There was a problem hiding this comment.
There is no need to extract the feature names here. When _validate_data is called, self.feature_names_in_ will be set. This can be passed into _parallel_build_trees directly:
diff --git a/sklearn/ensemble/_forest.py b/sklearn/ensemble/_forest.py
index de729264e3..44be556ca4 100644
--- a/sklearn/ensemble/_forest.py
+++ b/sklearn/ensemble/_forest.py
@@ -467,6 +467,7 @@ class BaseForest(MultiOutputMixin, BaseEnsemble, metaclass=ABCMeta):
verbose=self.verbose,
class_weight=self.class_weight,
n_samples_bootstrap=n_samples_bootstrap,
+ feature_names_in_=self.feature_names_in_,
)
for i, t in enumerate(trees)
)There was a problem hiding this comment.
Good one, let me try this.
There was a problem hiding this comment.
I tried your solution but a lot of the tests are failing
AttributeError: 'RandomForestClassifier' object has no attribute 'feature_names_in_'
Using the updated version of my branch all the tests are on green for ensamble/tests.
feature_names_in_ = []
if hasattr(X, "columns"):
feature_names_in_ = X.columns.to_list()
There was a problem hiding this comment.
Ah yes, feature_names_in_ is not always set. Here is an update suggestion:
diff --git a/sklearn/ensemble/_forest.py b/sklearn/ensemble/_forest.py
index 117e2e6016..e28b117e07 100644
--- a/sklearn/ensemble/_forest.py
+++ b/sklearn/ensemble/_forest.py
@@ -481,7 +481,7 @@ class BaseForest(MultiOutputMixin, BaseEnsemble, metaclass=ABCMeta):
verbose=self.verbose,
class_weight=self.class_weight,
n_samples_bootstrap=n_samples_bootstrap,
- feature_names_in_=feature_names_in_,
+ feature_names_in_=getattr(self, "feature_names_in_", None),
)
for i, t in enumerate(trees)
)
There was a problem hiding this comment.
Cheers, nice one. I implemented your suggestion and all tests are on green for sklearn/ensamble/tests.
yeah, I'd hold until we find a consensus there. But thanks for the PR. |
Reference Issues/PRs
Example: Fixes #26140
What does this implement/fix? Explain your changes.
In case of using a pandas DataFrame as the X variable type, every tree inside the forest will set its
feature_names_in_attribute equal the list of the columns of set pandas DataFrameAny other comments?