You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In scikit-learn, estimators have an _estimator_type attribute that indicates whether they are a classifier, regressor, or clusterer, etc.. This attribute is necessary for some scikit-learn API contracts, particularly around cross validation and meta-modeling. See here for the documentation.
For better compatibility with scikit-learn and higher-level tools enforcing the same contract, it would be useful if cuML estimators had this attribute as well. scikit-learn implements these in their Mixin classes. Because not all cuML estimators inherit from Mixins currently, perhaps these could be added in the estimator classes themselves as a start, and then handled via Mixin classes if/when additional estimators inherit from them.
Note that the Mixin classes already implement this attribute, so another approach would be to update estimators to inherit from the existing Mixins.
Perhaps these Mixins could be moved out of the metrics.base namespace and into the general common.base namespace? This would be consistent with sklearn
In scikit-learn, estimators have an
_estimator_type
attribute that indicates whether they are a classifier, regressor, or clusterer, etc.. This attribute is necessary for some scikit-learn API contracts, particularly around cross validation and meta-modeling. See here for the documentation.For better compatibility with scikit-learn and higher-level tools enforcing the same contract, it would be useful if cuML estimators had this attribute as well. scikit-learn implements these in their
Mixin
classes. Because not all cuML estimators inherit from Mixins currently, perhaps these could be added in the estimator classes themselves as a start, and then handled via Mixin classes if/when additional estimators inherit from them.Maybe something like:
The text was updated successfully, but these errors were encountered: