Merge pull request #121 from antoinedemathelin/master

fix: new sklearn version compatibility

Author: antoinedemathelin

adapt/instance_based/_iwc.py

@@ -124,7 +124,7 @@ def fit_weights(self, Xs, Xt, warm_start=False, **kwargs):
 
         if (not warm_start) or (not hasattr(self, "classifier_")):
             if self.classifier is None:
-                self.classifier_ = LogisticRegression(penalty="none")
+                self.classifier_ = LogisticRegression()
             else:
                 self.classifier_ = check_estimator(self.classifier,
                                                    copy=True,

adapt/instance_based/_tradaboost.py

@@ -320,7 +320,10 @@ def _boost(self, iboost, Xs, ys, Xt, yt,
 
         if not isinstance(self, TrAdaBoostR2):
             if isinstance(estimator, BaseEstimator):
-                ohe = OneHotEncoder(sparse=False)
+                try:
+                    ohe = OneHotEncoder(sparse=False)
+                except:
+                    ohe = OneHotEncoder(sparse_output=False)
                 ohe.fit(y.reshape(-1, 1))
                 ys = ohe.transform(ys)
                 yt = ohe.transform(yt)

tests/test_regular.py

@@ -56,7 +56,7 @@ def test_setup():
     lr.fit(Xs, ys_reg)
     assert np.abs(lr.coef_[0][0] - 10) < 1
 
-    lr = LogisticRegression(penalty='none', solver='lbfgs')
+    lr = LogisticRegression(penalty=None, solver='lbfgs')
     lr.fit(Xs, ys_classif)
     assert (lr.predict(Xt) == yt_classif.ravel()).sum() < 70
 
@@ -116,7 +116,7 @@ def test_regularlr_error():
 
 def test_regularlc_fit():
     np.random.seed(0)
-    lr = LogisticRegression(penalty='none', solver='lbfgs')
+    lr = LogisticRegression(penalty=None, solver='lbfgs')
     lr.fit(Xs, ys_classif)
     model = RegularTransferLC(lr, lambda_=0)
     model.fit(Xt, yt_classif)
@@ -139,7 +139,7 @@ def test_regularlc_multiclass():
     y = np.zeros(len(X))
     y[X[:, :2].sum(1)<0] = 1
     y[X[:, 3:].sum(1)>0] = 2
-    lr = LogisticRegression(penalty='none', solver='lbfgs')
+    lr = LogisticRegression(penalty=None, solver='lbfgs')
     lr.fit(X, y)
     model = RegularTransferLC(lr, lambda_=1.)
     model.fit(X, y)
@@ -193,7 +193,7 @@ def test_clone():
     new_model.predict(Xs);
     assert model is not new_model
 
-    lr = LogisticRegression(penalty='none', solver='lbfgs')
+    lr = LogisticRegression(penalty=None, solver='lbfgs')
     lr.fit(Xs, ys)
     model = RegularTransferLC(lr, lambda_=1.)
     model.fit(Xs, ys)

tests/test_tradaboost.py

@@ -36,7 +36,7 @@
 
 def test_tradaboost_fit():
     np.random.seed(0)
-    model = TrAdaBoost(LogisticRegression(penalty='none',
+    model = TrAdaBoost(LogisticRegression(penalty=None,
                        solver='lbfgs'),
                        n_estimators=20)
     model.fit(Xs, ys_classif, Xt=Xt[:10], yt=yt_classif[:10])
@@ -111,7 +111,7 @@ def test_twostagetradaboostr2_fit():
 
 def test_tradaboost_deepcopy():
     np.random.seed(0)
-    model = TrAdaBoost(LogisticRegression(penalty='none',
+    model = TrAdaBoost(LogisticRegression(penalty=None,
                        solver='lbfgs'),
                        n_estimators=20)
     model.fit(Xs, ys_classif, Xt=Xt[:10], yt=yt_classif[:10])
@@ -124,7 +124,7 @@ def test_tradaboost_multiclass():
     np.random.seed(0)
     X = np.random.randn(10, 3)
     y = np.random.choice(3, 10)
-    model = TrAdaBoost(LogisticRegression(penalty='none',
+    model = TrAdaBoost(LogisticRegression(penalty=None,
                        solver='lbfgs'), Xt=X, yt=y,
                        n_estimators=20)
     model.fit(X, y)
@@ -177,13 +177,13 @@ def test_tradaboost_above_05():
 
 def test_tradaboost_lr():
     np.random.seed(0)
-    model = TrAdaBoost(LogisticRegression(penalty='none'),
+    model = TrAdaBoost(LogisticRegression(penalty=None),
                          Xt=Xt[:10], yt=yt_classif[:10], 
                          n_estimators=20, lr=.1)
     model.fit(Xs, ys_classif)
     err1 = model.estimator_errors_
 
-    model = TrAdaBoost(LogisticRegression(penalty='none'),
+    model = TrAdaBoost(LogisticRegression(penalty=None),
                          Xt=Xt[:10], yt=yt_classif[:10], 
                          n_estimators=20, lr=2.)
     model.fit(Xs, ys_classif)