Initial

skprox/linear_model/_logistic.py

@@ -126,7 +126,7 @@ def fit(self, X, y, sample_weight=None):
         n_samples, n_features = X.shape
         if y.ndim == 1:
             self.ndim = 1
-            self.dims = (n_features,)
+            self.dims = (n_features,1)
         else:
             self.ndim = y.shape[1]
             self.dims = (n_features, y.shape[1])

skprox/test/test_logistic.py

@@ -1,24 +1,23 @@
 import numpy as np
-
+from sklearn.datasets import make_classification
+from sklearn.model_selection import train_test_split
 from skprox.linear_model import RegularisedLogisticRegression
 
-
-def test_L0_logistic():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
-
-    X, y, t = make_regression(
+X, y = make_classification(
         n_samples=100,
         n_features=200,
         n_informative=10,
-        n_targets=5,
         random_state=1,
-        coef=True,
     )
-    x_train, x_test, y_train, y_test = train_test_split(
+
+x_train, x_test, y_train, y_test = train_test_split(
         X, y, test_size=0.2, random_state=0
     )
-    reg = RegularisedLogisticRegression(proximal="L0", sigma=0.1, max_iter=10000)
+
+def test_L0_logistic():
+
+
+    reg = RegularisedLogisticRegression(proximal="L0", sigma=0.01, max_iter=10000)
     reg.fit(x_train, y_train)
     # test L0 has some zero coefficients
     assert np.sum(reg.coef_ == 0) > 0
@@ -27,28 +26,14 @@ def test_L0_logistic():
     assert np.sum(reg.coef_ != 0) > 0
 
     # test L0 has the same number of coefficients as the number of targets
-    assert reg.coef_.shape[0] == 5
+    assert reg.coef_.shape[0] == 1
 
     # test L0 has the same number of coefficients as the number of features
     assert reg.coef_.shape[1] == 200
 
 
 def test_L1_regression():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
-
-    X, y, t = make_regression(
-        n_samples=100,
-        n_features=200,
-        n_informative=10,
-        n_targets=5,
-        random_state=1,
-        coef=True,
-    )
-    x_train, x_test, y_train, y_test = train_test_split(
-        X, y, test_size=0.2, random_state=0
-    )
-    reg = RegularisedLogisticRegression(proximal="L1", sigma=0.1, max_iter=10000)
+    reg = RegularisedLogisticRegression(proximal="L1", sigma=0.01, max_iter=10000)
     reg.fit(x_train, y_train)
     # test L1 has some zero coefficients
     assert np.sum(reg.coef_ == 0) > 0
@@ -57,54 +42,26 @@ def test_L1_regression():
     assert np.sum(reg.coef_ != 0) > 0
 
     # test L1 has the same number of coefficients as the number of targets
-    assert reg.coef_.shape[0] == 5
+    assert reg.coef_.shape[0] == 1
 
     # test L1 has the same number of coefficients as the number of features
     assert reg.coef_.shape[1] == 200
 
 
 def test_nuclearball_regression():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
-
-    X, y, t = make_regression(
-        n_samples=100,
-        n_features=200,
-        n_informative=10,
-        n_targets=5,
-        random_state=1,
-        coef=True,
-    )
-    x_train, x_test, y_train, y_test = train_test_split(
-        X, y, test_size=0.2, random_state=0
-    )
     reg = RegularisedLogisticRegression(proximal="NuclearBall", radius=10, max_iter=10000)
     reg.fit(x_train, y_train)
 
     # test nuclearball has norm of coefficients about the same as the radius
     assert np.isclose(np.linalg.norm(reg.coef_), 10, atol=0.1)
 
     # test nuclear has the same number of coefficients as the number of targets
-    assert reg.coef_.shape[0] == 5
+    assert reg.coef_.shape[0] == 1
 
     # test nuclear has the same number of coefficients as the number of features
     assert reg.coef_.shape[1] == 200
 
 def test_gridsearch():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
-
-    X, y, t = make_regression(
-        n_samples=100,
-        n_features=200,
-        n_informative=10,
-        n_targets=5,
-        random_state=1,
-        coef=True,
-    )
-    x_train, x_test, y_train, y_test = train_test_split(
-        X, y, test_size=0.2, random_state=0
-    )
     from sklearn.model_selection import GridSearchCV
 
     reg = GridSearchCV(
@@ -115,7 +72,7 @@ def test_gridsearch():
     reg.fit(x_train, y_train)
 
     # test gridsearch has the same number of coefficients as the number of targets
-    assert reg.best_estimator_.coef_.shape[0] == 5
+    assert reg.best_estimator_.coef_.shape[0] == 1
 
     # test gridsearch has the same number of coefficients as the number of features
     assert reg.best_estimator_.coef_.shape[1] == 200

skprox/test/test_regression.py

@@ -1,23 +1,23 @@
 import numpy as np
-
+from sklearn.datasets import make_regression
+from sklearn.model_selection import train_test_split
 from skprox.linear_model import RegularisedLinearRegression
 
+X, y, t = make_regression(
+    n_samples=100,
+    n_features=200,
+    n_informative=10,
+    n_targets=5,
+    random_state=1,
+    coef=True,
+)
 
-def test_L0_regression():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
+x_train, x_test, y_train, y_test = train_test_split(
+    X, y, test_size=0.2, random_state=0
+)
 
-    X, y, t = make_regression(
-        n_samples=100,
-        n_features=200,
-        n_informative=10,
-        n_targets=5,
-        random_state=1,
-        coef=True,
-    )
-    x_train, x_test, y_train, y_test = train_test_split(
-        X, y, test_size=0.2, random_state=0
-    )
+
+def test_L0_logistic():
     reg = RegularisedLinearRegression(proximal="L0", sigma=0.1, max_iter=10000)
     reg.fit(x_train, y_train)
     # test L0 has some zero coefficients
@@ -34,20 +34,6 @@ def test_L0_regression():
 
 
 def test_L1_regression():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
-
-    X, y, t = make_regression(
-        n_samples=100,
-        n_features=200,
-        n_informative=10,
-        n_targets=5,
-        random_state=1,
-        coef=True,
-    )
-    x_train, x_test, y_train, y_test = train_test_split(
-        X, y, test_size=0.2, random_state=0
-    )
     reg = RegularisedLinearRegression(proximal="L1", sigma=0.1, max_iter=10000)
     reg.fit(x_train, y_train)
     # test L1 has some zero coefficients
@@ -64,20 +50,6 @@ def test_L1_regression():
 
 
 def test_nuclearball_regression():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
-
-    X, y, t = make_regression(
-        n_samples=100,
-        n_features=200,
-        n_informative=10,
-        n_targets=5,
-        random_state=1,
-        coef=True,
-    )
-    x_train, x_test, y_train, y_test = train_test_split(
-        X, y, test_size=0.2, random_state=0
-    )
     reg = RegularisedLinearRegression(proximal="NuclearBall", radius=10, max_iter=10000)
     reg.fit(x_train, y_train)
 
@@ -90,26 +62,13 @@ def test_nuclearball_regression():
     # test nuclear has the same number of coefficients as the number of features
     assert reg.coef_.shape[1] == 200
 
+
 def test_gridsearch():
-    from sklearn.datasets import make_regression
-    from sklearn.model_selection import train_test_split
-
-    X, y, t = make_regression(
-        n_samples=100,
-        n_features=200,
-        n_informative=10,
-        n_targets=5,
-        random_state=1,
-        coef=True,
-    )
-    x_train, x_test, y_train, y_test = train_test_split(
-        X, y, test_size=0.2, random_state=0
-    )
     from sklearn.model_selection import GridSearchCV
 
     reg = GridSearchCV(
         RegularisedLinearRegression(max_iter=1000),
-        param_grid={"proximal": ["L0", "L1"], "sigma": [1e-1,1,10]},
+        param_grid={"proximal": ["L0", "L1"], "sigma": [1e-1, 1, 10]},
         cv=5,
     )
     reg.fit(x_train, y_train)