fix behavior when sample weight is None

mlxtend/regressor/stacking_cv_regression.py

@@ -195,11 +195,18 @@ def fit(self, X, y, groups=None, sample_weight=None):
             meta_features = sparse.hstack((X, meta_features))
         else:
             meta_features = np.hstack((X, meta_features))
-        self.meta_regr_.fit(meta_features, y, sample_weight=sample_weight)
+
+        if sample_weight is None:
+            self.meta_regr_.fit(meta_features, y)
+        else:
+            self.meta_regr_.fit(meta_features, y, sample_weight=sample_weight)
 
         # Retrain base models on all data
         for regr in self.regr_:
-            regr.fit(X, y, sample_weight=sample_weight)
+            if sample_weight is None:
+                regr.fit(X, y)
+            else:
+                regr.fit(X, y, sample_weight=sample_weight)
 
         return self
 

mlxtend/regressor/stacking_regression.py

@@ -146,7 +146,10 @@ def fit(self, X, y, sample_weight=None):
             if self.verbose > 1:
                 print(_name_estimators((regr,))[0][1])
 
-            regr.fit(X, y, sample_weight=sample_weight)
+            if sample_weight is None:
+                regr.fit(X, y)
+            else:
+                regr.fit(X, y, sample_weight=sample_weight)
 
         meta_features = self.predict_meta_features(X)
 
@@ -157,7 +160,11 @@ def fit(self, X, y, sample_weight=None):
             meta_features = sparse.hstack((X, meta_features))
         else:
             meta_features = np.hstack((X, meta_features))
-        self.meta_regr_.fit(meta_features, y, sample_weight=sample_weight)
+
+        if sample_weight is None:
+            self.meta_regr_.fit(meta_features, y)
+        else:
+            self.meta_regr_.fit(meta_features, y, sample_weight=sample_weight)
 
         # save meta-features for training data
         if self.store_train_meta_features:

mlxtend/regressor/tests/test_stacking_cv_regression.py

@@ -15,7 +15,6 @@
 from mlxtend.utils import assert_raises
 from sklearn.linear_model import LinearRegression
 from sklearn.linear_model import Ridge, Lasso
-from sklearn.neural_network import MLPRegressor
 from sklearn.svm import SVR
 from sklearn.model_selection import GridSearchCV, train_test_split, KFold
 from sklearn.base import clone
@@ -268,10 +267,13 @@ def test_sample_weight():
     stack = StackingCVRegressor(regressors=[svr_lin, lr, ridge],
                                 meta_regressor=svr_rbf,
                                 cv=KFold(4, shuffle=True, random_state=7))
-    stack.fit(X1, y, sample_weight=w).predict(X1)
+    pred1 = stack.fit(X1, y, sample_weight=w).predict(X1)
     mse = 0.21  # 0.20770
     got = np.mean((stack.predict(X1) - y) ** 2)
     assert round(got, 2) == mse, "Expected %.2f, but got %.5f" % (mse, got)
+    pred2 = stack.fit(X1, y).predict(X1)
+    maxdiff = np.max(np.abs(pred1 - pred2))
+    assert maxdiff > 1e-3, "max diff is %.4f" % maxdiff
 
 
 def test_weight_ones():
@@ -299,15 +301,30 @@ def test_unsupported_regressor():
     svr_rbf = SVR(kernel='rbf')
     stack = StackingCVRegressor(regressors=[svr_lin, lr, ridge, lasso],
                                 meta_regressor=svr_rbf)
-    stack.fit(X1, y).predict(X1)
+    stack.fit(X1, y, sample_weight=w).predict(X1)
 
 
 @raises(TypeError)
 def test_unsupported_meta_regressor():
     lr = LinearRegression()
     svr_lin = SVR(kernel='linear')
     ridge = Ridge(random_state=1)
-    mlp = MLPRegressor()
+    lasso = Lasso()
+    stack = StackingCVRegressor(regressors=[svr_lin, lr, ridge],
+                                meta_regressor=lasso)
+    stack.fit(X1, y, sample_weight=w).predict(X1)
+
+
+def test_weight_unsupported_with_no_weight():
+    # should be okay since we do not pass weight
+    lr = LinearRegression()
+    svr_lin = SVR(kernel='linear')
+    ridge = Ridge(random_state=1)
+    lasso = Lasso()
+    stack = StackingCVRegressor(regressors=[svr_lin, lr, lasso],
+                                meta_regressor=ridge)
+    stack.fit(X1, y).predict(X1)
+
     stack = StackingCVRegressor(regressors=[svr_lin, lr, ridge],
-                                meta_regressor=mlp)
+                                meta_regressor=lasso)
     stack.fit(X1, y).predict(X1)

mlxtend/regressor/tests/test_stacking_regression.py

@@ -11,7 +11,6 @@
 from sklearn.linear_model import LinearRegression
 from sklearn.linear_model import Ridge
 from sklearn.linear_model import Lasso
-from sklearn.neural_network import MLPRegressor
 from sklearn.svm import SVR
 from sklearn.ensemble import RandomForestRegressor
 from sklearn.model_selection import GridSearchCV
@@ -79,10 +78,14 @@ def test_sample_weight():
     svr_rbf = SVR(kernel='rbf')
     stregr = StackingRegressor(regressors=[svr_lin, lr, ridge],
                                meta_regressor=svr_rbf)
-    stregr.fit(X1, y, sample_weight=w).predict(X1)
+    pred1 = stregr.fit(X1, y, sample_weight=w).predict(X1)
     mse = 0.22
     got = np.mean((stregr.predict(X1) - y) ** 2)
     assert round(got, 2) == mse
+    # make sure that this is not equivalent to the model with no weight
+    pred2 = stregr.fit(X1, y).predict(X1)
+    maxdiff = np.max(np.abs(pred1 - pred2))
+    assert maxdiff > 1e-3, "max diff is %.4f" % maxdiff
 
 
 def test_weight_ones():
@@ -93,15 +96,14 @@ def test_weight_ones():
     svr_rbf = SVR(kernel='rbf')
     stregr = StackingRegressor(regressors=[svr_lin, lr, ridge],
                                meta_regressor=svr_rbf)
-
-    stregr.fit(X1, y, sample_weight=np.ones(40)).predict(X1)
-    mse = 0.21
-    got = np.mean((stregr.predict(X1) - y) ** 2)
-    assert round(got, 2) == mse
+    pred1 = stregr.fit(X1, y).predict(X1)
+    pred2 = stregr.fit(X1, y, sample_weight=np.ones(40)).predict(X1)
+    maxdiff = np.max(np.abs(pred1 - pred2))
+    assert maxdiff < 1e-3, "max diff is %.4f" % maxdiff
 
 
 @raises(TypeError)
-def test_unsupported_regressor():
+def test_weight_unsupported_regressor():
     # including regressor that does not support
     # sample_weight should raise error
     lr = LinearRegression()
@@ -111,19 +113,37 @@ def test_unsupported_regressor():
     lasso = Lasso(random_state=1)
     stregr = StackingRegressor(regressors=[svr_lin, lr, ridge, lasso],
                                meta_regressor=svr_rbf)
-    stregr.fit(X1, y, w).predict(X1)
+    stregr.fit(X1, y, sample_weight=w).predict(X1)
 
 
 @raises(TypeError)
-def test_unsupported_meta_regressor():
-    # meta regressor that does not support sample_weight should raise error
+def test_weight_unsupported_meta():
+    # meta regressor with no support for
+    # sample_weight should raise error
+    lr = LinearRegression()
+    svr_lin = SVR(kernel='linear')
+    ridge = Ridge(random_state=1)
+    lasso = Lasso(random_state=1)
+    stregr = StackingRegressor(regressors=[svr_lin, lr, ridge],
+                               meta_regressor=lasso)
+    stregr.fit(X1, y, sample_weight=w).predict(X1)
+
+
+def test_weight_unsupported_with_no_weight():
+    # pass no weight to regressors with no weight support
+    # should not be a problem
     lr = LinearRegression()
     svr_lin = SVR(kernel='linear')
     ridge = Ridge(random_state=1)
-    mlp = MLPRegressor()
+    svr_rbf = SVR(kernel='rbf')
+    lasso = Lasso(random_state=1)
+    stregr = StackingRegressor(regressors=[svr_lin, lr, ridge, lasso],
+                               meta_regressor=svr_rbf)
+    stregr.fit(X1, y).predict(X1)
+
     stregr = StackingRegressor(regressors=[svr_lin, lr, ridge],
-                               meta_regressor=mlp)
-    stregr.fit(X1, y, w).predict(X1)
+                               meta_regressor=lasso)
+    stregr.fit(X1, y).predict(X1)
 
 
 def test_gridsearch():