Fixing sklearn error when using RandomizedSearchCV

Signed-off-by: Luis França <[email protected]>

python/interpret-core/interpret/glassbox/linear.py

@@ -10,12 +10,12 @@
 from abc import abstractmethod
 from sklearn.base import is_classifier
 import numpy as np
-from sklearn.base import ClassifierMixin, RegressorMixin
+from sklearn.base import BaseEstimator, ClassifierMixin, RegressorMixin
 from sklearn.linear_model import LogisticRegression as SKLogistic
 from sklearn.linear_model import Lasso as SKLinear
 
 
-class BaseLinear:
+class BaseLinear(BaseEstimator):
     """ Base linear model.
 
     Currently wrapper around linear models in scikit-learn.
@@ -43,11 +43,26 @@ def __init__(
         self.linear_class = linear_class
         self.kwargs = kwargs
 
+        for key, value in self.kwargs.items():
+            setattr(self, key, value)
+
     @abstractmethod
     def _model(self):
         # This method should be overridden.
         return None
 
+    # get_params and set_params are usually inherited from BaseEstimator, but they will 
+    # fail here due to the **kwargs in the __init__. Therefore, we implement them.
+    def get_params(self, deep = True):
+        return {param: getattr(self, param)
+                    for param in self.kwargs}
+
+    def set_params(self, **parameters):
+        for parameter, value in parameters.items():
+            setattr(self, parameter, value)
+
+        return self
+
     def fit(self, X, y):
         """ Fits model to provided instances.
 

python/interpret-core/interpret/glassbox/test/test_linear.py

@@ -5,6 +5,7 @@
 from sklearn.datasets import load_breast_cancer, load_boston
 from sklearn.linear_model import LogisticRegression as SKLogistic
 from sklearn.linear_model import Lasso as SKLinear
+from sklearn.model_selection import RandomizedSearchCV
 import numpy as np
 
 
@@ -38,6 +39,35 @@ def test_linear_regression():
     assert global_viz is not None
 
 
+def test_linear_regression_sklearn_compatibility():
+    boston = load_boston()
+    X, y = boston.data, boston.target
+
+    distributions = {
+        'max_iter': [250, 500],
+        'alpha': [0.1 , 0.25, 0.5, 1]
+    }
+
+    sk_lr = SKLinear()
+    our_lr = LinearRegression()
+
+    search_sk =  RandomizedSearchCV(estimator = sk_lr,
+                                    param_distributions = distributions,
+                                    random_state = 2022)
+
+    search_our = RandomizedSearchCV(estimator = our_lr,
+                                    param_distributions = distributions,
+                                    random_state = 2022)
+
+    search_sk.fit(X, y)
+    search_our.fit(X, y)
+
+    sk_pred = search_sk.predict(X)
+    our_pred = search_our.predict(X)
+
+    assert np.allclose(sk_pred, our_pred)
+
+
 def test_logistic_regression():
     cancer = load_breast_cancer()
     X, y = cancer.data, cancer.target
@@ -72,6 +102,35 @@ def test_logistic_regression():
     assert global_viz is not None
 
 
+def test_logistic_regression_sklearn_compatibility():
+    cancer = load_breast_cancer()
+    X, y = cancer.data, cancer.target
+
+    distributions = {
+        'penalty': ['l1', 'l2'],
+        'C': [1 , 0.5, 0.1, 0.05, 0.01]
+    }
+
+    sk_lr = SKLogistic()
+    our_lr = LogisticRegression()
+
+    search_sk =  RandomizedSearchCV(estimator = sk_lr,
+                                    param_distributions = distributions,
+                                    random_state = 2022)
+
+    search_our = RandomizedSearchCV(estimator = our_lr,
+                                    param_distributions = distributions,
+                                    random_state = 2022)
+
+    search_sk.fit(X, y)
+    search_our.fit(X, y)
+
+    sk_pred = search_sk.predict_proba(X)
+    our_pred = search_our.predict_proba(X)
+
+    assert np.allclose(sk_pred, our_pred)
+
+
 def test_sorting():
     cancer = load_breast_cancer()
     X, y = cancer.data, cancer.target