[MRG] Fix bug in graph lasso when n_features=2 (scikit-learn#13276)

* fix bug in graph lasso

* better fix

* add test for case n_features=2

* remove test for deprecated version

* update whats new

* note that the bug was a regression

Author: bellet

doc/whats_new/v0.20.rst

@@ -29,6 +29,13 @@ Changelog
   negative indexes in the columns list of the transformers.
   :issue:`12946` by :user:`Pierre Tallotte <pierretallotte>`.
 
+:mod:`sklearn.covariance`
+......................
+
+- |Fix| Fixed a regression in :func:`covariance.graphical_lasso` so that
+  the case `n_features=2` is handled correctly. :issue:`13276` by
+  :user:`Aurélien Bellet <bellet>`.
+
 :mod:`sklearn.decomposition`
 ............................
 

sklearn/covariance/graph_lasso_.py

@@ -204,7 +204,7 @@ def graphical_lasso(emp_cov, alpha, cov_init=None, mode='cd', tol=1e-4,
         # https://github.com/scikit-learn/scikit-learn/issues/4134
         d_gap = np.inf
         # set a sub_covariance buffer
-        sub_covariance = np.ascontiguousarray(covariance_[1:, 1:])
+        sub_covariance = np.copy(covariance_[1:, 1:], order='C')
         for i in range(max_iter):
             for idx in range(n_features):
                 # To keep the contiguous matrix `sub_covariance` equal to

sklearn/covariance/tests/test_graphical_lasso.py

@@ -85,6 +85,23 @@ def test_graphical_lasso_iris():
         assert_array_almost_equal(icov, icov_R)
 
 
+def test_graph_lasso_2D():
+    # Hard-coded solution from Python skggm package
+    # obtained by calling `quic(emp_cov, lam=.1, tol=1e-8)`
+    cov_skggm = np.array([[3.09550269, 1.186972],
+                         [1.186972, 0.57713289]])
+
+    icov_skggm = np.array([[1.52836773, -3.14334831],
+                          [-3.14334831,  8.19753385]])
+    X = datasets.load_iris().data[:, 2:]
+    emp_cov = empirical_covariance(X)
+    for method in ('cd', 'lars'):
+        cov, icov = graphical_lasso(emp_cov, alpha=.1, return_costs=False,
+                                    mode=method)
+        assert_array_almost_equal(cov, cov_skggm)
+        assert_array_almost_equal(icov, icov_skggm)
+
+
 def test_graphical_lasso_iris_singular():
     # Small subset of rows to test the rank-deficient case
     # Need to choose samples such that none of the variances are zero