Fix regression algorithms to give correct output dimensions (#1335)

* Added ignored_warnings file

* Use ignored_warnings file

* Test regressors with 1d, 1d as 2d and 2d targets

* Flake'd

* Fix broken relative imports to ignore_warnings

* Removed print and updated parameter type for tests

* Type import fix

Author: eddiebergman

autosklearn/pipeline/components/regression/adaboost.py

@@ -15,7 +15,7 @@ def __init__(self, n_estimators, learning_rate, loss, max_depth, random_state=No
         self.max_depth = max_depth
         self.estimator = None
 
-    def fit(self, X, Y):
+    def fit(self, X, y):
         import sklearn.ensemble
         import sklearn.tree
 
@@ -32,7 +32,11 @@ def fit(self, X, Y):
             loss=self.loss,
             random_state=self.random_state
         )
-        self.estimator.fit(X, Y)
+
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
+        self.estimator.fit(X, y)
         return self
 
     def predict(self, X):

autosklearn/pipeline/components/regression/ard_regression.py

@@ -22,8 +22,8 @@ def __init__(self, n_iter, tol, alpha_1, alpha_2, lambda_1, lambda_2,
         self.threshold_lambda = threshold_lambda
         self.fit_intercept = fit_intercept
 
-    def fit(self, X, Y):
-        import sklearn.linear_model
+    def fit(self, X, y):
+        from sklearn.linear_model import ARDRegression
 
         self.n_iter = int(self.n_iter)
         self.tol = float(self.tol)
@@ -34,20 +34,25 @@ def fit(self, X, Y):
         self.threshold_lambda = float(self.threshold_lambda)
         self.fit_intercept = check_for_bool(self.fit_intercept)
 
-        self.estimator = sklearn.linear_model.\
-            ARDRegression(n_iter=self.n_iter,
-                          tol=self.tol,
-                          alpha_1=self.alpha_1,
-                          alpha_2=self.alpha_2,
-                          lambda_1=self.lambda_1,
-                          lambda_2=self.lambda_2,
-                          compute_score=False,
-                          threshold_lambda=self.threshold_lambda,
-                          fit_intercept=True,
-                          normalize=False,
-                          copy_X=False,
-                          verbose=False)
-        self.estimator.fit(X, Y)
+        self.estimator = ARDRegression(
+            n_iter=self.n_iter,
+            tol=self.tol,
+            alpha_1=self.alpha_1,
+            alpha_2=self.alpha_2,
+            lambda_1=self.lambda_1,
+            lambda_2=self.lambda_2,
+            compute_score=False,
+            threshold_lambda=self.threshold_lambda,
+            fit_intercept=True,
+            normalize=False,
+            copy_X=False,
+            verbose=False
+        )
+
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
+        self.estimator.fit(X, y)
         return self
 
     def predict(self, X):

autosklearn/pipeline/components/regression/decision_tree.py

@@ -56,6 +56,10 @@ def fit(self, X, y, sample_weight=None):
             min_weight_fraction_leaf=self.min_weight_fraction_leaf,
             min_impurity_decrease=self.min_impurity_decrease,
             random_state=self.random_state)
+
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
         self.estimator.fit(X, y, sample_weight=sample_weight)
         return self
 

autosklearn/pipeline/components/regression/extra_trees.py

@@ -95,7 +95,10 @@ def iterative_fit(self, X, y, n_iter=1, refit=False):
             self.estimator.n_estimators = min(self.estimator.n_estimators,
                                               self.n_estimators)
 
-        self.estimator.fit(X, y,)
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
+        self.estimator.fit(X, y)
 
         return self
 

autosklearn/pipeline/components/regression/gaussian_process.py

@@ -12,7 +12,6 @@ def __init__(self, alpha, thetaL, thetaU, random_state=None):
         self.thetaU = thetaU
         self.random_state = random_state
         self.estimator = None
-        self.scaler = None
 
     def fit(self, X, y):
         import sklearn.gaussian_process
@@ -38,6 +37,9 @@ def fit(self, X, y):
             normalize_y=True
         )
 
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
         self.estimator.fit(X, y)
 
         return self

autosklearn/pipeline/components/regression/gradient_boosting.py

@@ -48,10 +48,7 @@ def get_current_iter(self):
         return self.estimator.n_iter_
 
     def iterative_fit(self, X, y, n_iter=2, refit=False):
-
-        """
-        Set n_iter=2 for the same reason as for SGD
-        """
+        """ Set n_iter=2 for the same reason as for SGD """
         import sklearn.ensemble
         from sklearn.experimental import enable_hist_gradient_boosting  # noqa
 
@@ -112,6 +109,9 @@ def iterative_fit(self, X, y, n_iter=2, refit=False):
             self.estimator.max_iter = min(self.estimator.max_iter,
                                           self.max_iter)
 
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
         self.estimator.fit(X, y)
 
         if (

autosklearn/pipeline/components/regression/k_nearest_neighbors.py

@@ -13,7 +13,7 @@ def __init__(self, n_neighbors, weights, p, random_state=None):
         self.p = p
         self.random_state = random_state
 
-    def fit(self, X, Y):
+    def fit(self, X, y):
         import sklearn.neighbors
 
         self.n_neighbors = int(self.n_neighbors)
@@ -24,7 +24,11 @@ def fit(self, X, Y):
                 n_neighbors=self.n_neighbors,
                 weights=self.weights,
                 p=self.p)
-        self.estimator.fit(X, Y)
+
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
+        self.estimator.fit(X, y)
         return self
 
     def predict(self, X):

autosklearn/pipeline/components/regression/liblinear_svr.py

@@ -23,7 +23,7 @@ def __init__(self, loss, epsilon, dual, tol, C, fit_intercept,
         self.random_state = random_state
         self.estimator = None
 
-    def fit(self, X, Y):
+    def fit(self, X, y):
         import sklearn.svm
 
         self.C = float(self.C)
@@ -42,7 +42,11 @@ def fit(self, X, Y):
                                                fit_intercept=self.fit_intercept,
                                                intercept_scaling=self.intercept_scaling,
                                                random_state=self.random_state)
-        self.estimator.fit(X, Y)
+
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
+        self.estimator.fit(X, y)
         return self
 
     def predict(self, X):

autosklearn/pipeline/components/regression/libsvm_svr.py

@@ -6,7 +6,6 @@
 from ConfigSpace.hyperparameters import UniformFloatHyperparameter, \
     UniformIntegerHyperparameter, CategoricalHyperparameter, \
     UnParametrizedHyperparameter
-
 from autosklearn.pipeline.components.base import AutoSklearnRegressionAlgorithm
 from autosklearn.pipeline.constants import DENSE, UNSIGNED_DATA, PREDICTIONS, SPARSE
 from autosklearn.util.common import check_for_bool, check_none
@@ -29,7 +28,7 @@ def __init__(self, kernel, C, epsilon, tol, shrinking, gamma=0.1,
         self.random_state = random_state
         self.estimator = None
 
-    def fit(self, X, Y):
+    def fit(self, X, y):
         import sklearn.svm
 
         # Calculate the size of the kernel cache (in MB) for sklearn's LibSVM. The cache size is
@@ -88,18 +87,35 @@ def fit(self, X, Y):
         )
         self.scaler = sklearn.preprocessing.StandardScaler(copy=True)
 
-        self.scaler.fit(Y.reshape((-1, 1)))
-        Y_scaled = self.scaler.transform(Y.reshape((-1, 1))).ravel()
-        self.estimator.fit(X, Y_scaled)
+        # Convert y to be at least 2d for the scaler
+        # [1,1,1] -> [[1], [1], [1]]
+        if y.ndim == 1:
+            y = y.reshape((-1, 1))
+
+        y_scaled = self.scaler.fit_transform(y)
+
+        # Flatten: [[0], [0], [0]] -> [0, 0, 0]
+        if y_scaled.ndim == 2 and y_scaled.shape[1] == 1:
+            y_scaled = y_scaled.flatten()
+
+        self.estimator.fit(X, y_scaled)
+
         return self
 
     def predict(self, X):
         if self.estimator is None:
             raise NotImplementedError
         if self.scaler is None:
             raise NotImplementedError
-        Y_pred = self.estimator.predict(X)
-        return self.scaler.inverse_transform(Y_pred)
+        y_pred = self.estimator.predict(X)
+
+        inverse = self.scaler.inverse_transform(y_pred)
+
+        # Flatten: [[0], [0], [0]] -> [0, 0, 0]
+        if inverse.ndim == 2 and inverse.shape[1] == 1:
+            inverse = inverse.flatten()
+
+        return inverse
 
     @staticmethod
     def get_properties(dataset_properties=None):

autosklearn/pipeline/components/regression/mlp.py

@@ -137,16 +137,36 @@ def iterative_fit(self, X, y, n_iter=2, refit=False):
                 # max_fun=self.max_fun
             )
             self.scaler = sklearn.preprocessing.StandardScaler(copy=True)
-            self.scaler.fit(y.reshape((-1, 1)))
+
+            # Convert y to be at least 2d for the StandardScaler
+            # [1,1,1] -> [[1], [1], [1]]
+            if y.ndim == 1:
+                y = y.reshape((-1, 1))
+
+            self.scaler.fit(y)
         else:
             new_max_iter = min(self.max_iter - self.estimator.n_iter_, n_iter)
             self.estimator.max_iter = new_max_iter
 
-        Y_scaled = self.scaler.transform(y.reshape((-1, 1))).ravel()
-        self.estimator.fit(X, Y_scaled)
-        if self.estimator.n_iter_ >= self.max_iter or \
-                self.estimator._no_improvement_count > self.n_iter_no_change:
+        # Convert y to be at least 2d for the scaler
+        # [1,1,1] -> [[1], [1], [1]]
+        if y.ndim == 1:
+            y = y.reshape((-1, 1))
+
+        y_scaled = self.scaler.transform(y)
+
+        # Flatten: [[0], [0], [0]] -> [0, 0, 0]
+        if y_scaled.ndim == 2 and y_scaled.shape[1] == 1:
+            y_scaled = y_scaled.flatten()
+
+        self.estimator.fit(X, y_scaled)
+
+        if (
+            self.estimator.n_iter_ >= self.max_iter
+            or self.estimator._no_improvement_count > self.n_iter_no_change
+        ):
             self._fully_fit = True
+
         return self
 
     def configuration_fully_fitted(self):
@@ -160,8 +180,16 @@ def configuration_fully_fitted(self):
     def predict(self, X):
         if self.estimator is None:
             raise NotImplementedError
-        Y_pred = self.estimator.predict(X)
-        return self.scaler.inverse_transform(Y_pred)
+
+        y_pred = self.estimator.predict(X)
+
+        inverse = self.scaler.inverse_transform(y_pred)
+
+        # Flatten: [[0], [0], [0]] -> [0, 0, 0]
+        if inverse.ndim == 2 and inverse.shape[1] == 1:
+            inverse = inverse.flatten()
+
+        return inverse
 
     @staticmethod
     def get_properties(dataset_properties=None):

autosklearn/pipeline/components/regression/random_forest.py

@@ -85,6 +85,9 @@ def iterative_fit(self, X, y, n_iter=1, refit=False):
             self.estimator.n_estimators = min(self.estimator.n_estimators,
                                               self.n_estimators)
 
+        if y.ndim == 2 and y.shape[1] == 1:
+            y = y.flatten()
+
         self.estimator.fit(X, y)
         return self
 

autosklearn/pipeline/components/regression/sgd.py

@@ -90,17 +90,36 @@ def iterative_fit(self, X, y, n_iter=2, refit=False):
                                           warm_start=True)
 
             self.scaler = sklearn.preprocessing.StandardScaler(copy=True)
-            self.scaler.fit(y.reshape((-1, 1)))
-            Y_scaled = self.scaler.transform(y.reshape((-1, 1))).ravel()
-            self.estimator.fit(X, Y_scaled)
+
+            if y.ndim == 1:
+                y = y.reshape((-1, 1))
+
+            y_scaled = self.scaler.fit_transform(y)
+
+            # Flatten: [[0], [0], [0]] -> [0, 0, 0]
+            if y_scaled.ndim == 2 and y_scaled.shape[1] == 1:
+                y_scaled = y_scaled.flatten()
+
+            self.estimator.fit(X, y_scaled)
             self.n_iter_ = self.estimator.n_iter_
         else:
             self.estimator.max_iter += n_iter
             self.estimator.max_iter = min(self.estimator.max_iter, self.max_iter)
-            Y_scaled = self.scaler.transform(y.reshape((-1, 1))).ravel()
+
+            # Convert y to be at least 2d for the scaler
+            # [1,1,1] -> [[1], [1], [1]]
+            if y.ndim == 1:
+                y = y.reshape((-1, 1))
+
+            y_scaled = self.scaler.transform(y)
+
+            # Flatten: [[0], [0], [0]] -> [0, 0, 0]
+            if y_scaled.ndim == 2 and y_scaled.shape[1] == 1:
+                y_scaled = y_scaled.flatten()
+
             self.estimator._validate_params()
             self.estimator._partial_fit(
-                X, Y_scaled,
+                X, y_scaled,
                 alpha=self.estimator.alpha,
                 C=1.0,
                 loss=self.estimator.loss,