Fix RandomState in the tests of linear model. (scikit-learn#7881)

sklearn/linear_model/tests/test_coordinate_descent.py

@@ -437,8 +437,9 @@ def test_enet_multitarget():
 
 
 def test_multioutput_enetcv_error():
- X = np.random.randn(10, 2)
- y = np.random.randn(10, 2)
+ rng = np.random.RandomState(0)
+ X = rng.randn(10, 2)
+ y = rng.randn(10, 2)
  clf = ElasticNetCV()
  assert_raises(ValueError, clf.fit, X, y)
 

sklearn/linear_model/tests/test_least_angle.py

@@ -107,6 +107,7 @@ def test_collinearity():
  [2., 2., 0.],
  [1., 1., 0]])
  y = np.array([1., 0., 0])
+ rng = np.random.RandomState(0)
 
  f = ignore_warnings
  _, _, coef_path_ = f(linear_model.lars_path)(X, y, alpha_min=0.01)
@@ -115,7 +116,7 @@ def test_collinearity():
  assert_less((residual ** 2).sum(), 1.) # just make sure it's bounded
 
  n_samples = 10
- X = np.random.rand(n_samples, 5)
+ X = rng.rand(n_samples, 5)
  y = np.zeros(n_samples)
  _, _, coef_path_ = linear_model.lars_path(X, y, Gram='auto', copy_X=False,
  copy_Gram=False, alpha_min=0.,

sklearn/linear_model/tests/test_ransac.py

@@ -54,8 +54,9 @@ def is_data_valid(X, y):
  assert_equal(y.shape[0], 2)
  return False
 
- X = np.random.rand(10, 2)
- y = np.random.rand(10, 1)
+ rng = np.random.RandomState(0)
+ X = rng.rand(10, 2)
+ y = rng.rand(10, 1)
 
  base_estimator = LinearRegression()
  ransac_estimator = RANSACRegressor(base_estimator, min_samples=2,

sklearn/linear_model/tests/test_sgd.py

@@ -611,7 +611,8 @@ def test_wrong_class_weight_format(self):
  def test_weights_multiplied(self):
  # Tests that class_weight and sample_weight are multiplicative
  class_weights = {1: .6, 2: .3}
- sample_weights = np.random.random(Y4.shape[0])
+ rng = np.random.RandomState(0)
+ sample_weights = rng.random_sample(Y4.shape[0])
  multiplied_together = np.copy(sample_weights)
  multiplied_together[Y4 == 1] *= class_weights[1]
  multiplied_together[Y4 == 2] *= class_weights[2]
@@ -960,6 +961,7 @@ def test_sgd_least_squares_fit(self):
  def test_sgd_epsilon_insensitive(self):
  xmin, xmax = -5, 5
  n_samples = 100
+ rng = np.random.RandomState(0)
  X = np.linspace(xmin, xmax, n_samples).reshape(n_samples, 1)
 
  # simple linear function without noise
@@ -973,8 +975,7 @@ def test_sgd_epsilon_insensitive(self):
  assert_true(score > 0.99)
 
  # simple linear function with noise
- y = 0.5 * X.ravel() \
- + np.random.randn(n_samples, 1).ravel()
+ y = 0.5 * X.ravel() + rng.randn(n_samples, 1).ravel()
 
  clf = self.factory(loss='epsilon_insensitive', epsilon=0.01,
  alpha=0.1, n_iter=20,
@@ -1012,7 +1013,7 @@ def test_elasticnet_convergence(self):
 
  n_samples, n_features = 1000, 5
  rng = np.random.RandomState(0)
- X = np.random.randn(n_samples, n_features)
+ X = rng.randn(n_samples, n_features)
  # ground_truth linear model that generate y from X and to which the
  # models should converge if the regularizer would be set to 0.0
  ground_truth_coef = rng.randn(n_features)