scikit-learn · jnothman · Mar 6, 2019 · Jun 15, 2018 · Jun 15, 2018 · Jul 25, 2018
diff --git a/benchmarks/bench_plot_lasso_path.py b/benchmarks/bench_plot_lasso_path.py
@@ -9,7 +9,7 @@
 
 import numpy as np
 
-from sklearn.linear_model import lars_path
+from sklearn.linear_model import lars_path, lars_path_gram
 from sklearn.linear_model import lasso_path
 from sklearn.datasets.samples_generator import make_regression
 
@@ -45,7 +45,7 @@ def compute_bench(samples_range, features_range):
             tstart = time()
             G = np.dot(X.T, X)  # precomputed Gram matrix
             Xy = np.dot(X.T, y)
-            lars_path(X, y, Xy=Xy, Gram=G, method='lasso')
+            lars_path_gram(Xy=Xy, Gram=G, n_samples=y.size, method='lasso')
             delta = time() - tstart
             print("%0.3fs" % delta)
             results['lars_path (with Gram)'].append(delta)

diff --git a/benchmarks/bench_plot_omp_lars.py b/benchmarks/bench_plot_omp_lars.py
@@ -9,7 +9,7 @@
 
 import numpy as np
 
-from sklearn.linear_model import lars_path, orthogonal_mp
+from sklearn.linear_model import lars_path, lars_path_gram, orthogonal_mp
 from sklearn.datasets.samples_generator import make_sparse_coded_signal
 
 
@@ -58,7 +58,8 @@ def compute_bench(samples_range, features_range):
             tstart = time()
             G = np.dot(X.T, X)  # precomputed Gram matrix
             Xy = np.dot(X.T, y)
-            lars_path(X, y, Xy=Xy, Gram=G, max_iter=n_informative)
+            lars_path_gram(Xy=Xy, Gram=G, n_samples=y.size,
+                           max_iter=n_informative)
             delta = time() - tstart
             print("%0.3fs" % delta)
             lars_gram[i_f, i_s] = delta

diff --git a/doc/modules/classes.rst b/doc/modules/classes.rst
@@ -756,6 +756,7 @@ Kernels:
 
    linear_model.enet_path
    linear_model.lars_path
+   linear_model.lars_path_gram
    linear_model.lasso_path
    linear_model.orthogonal_mp
    linear_model.orthogonal_mp_gram

diff --git a/doc/modules/linear_model.rst b/doc/modules/linear_model.rst
@@ -456,7 +456,7 @@ The disadvantages of the LARS method include:
     Statistics article.
 
 The LARS model can be used using estimator :class:`Lars`, or its
-low-level implementation :func:`lars_path`.
+low-level implementation :func:`lars_path` or :func:`lars_path_gram`.
 
 
 LARS Lasso
@@ -489,7 +489,8 @@ function of the norm of its coefficients.
 
 The Lars algorithm provides the full path of the coefficients along
 the regularization parameter almost for free, thus a common operation
-consist of retrieving the path with function :func:`lars_path`
+consist of retrieving the path with function :func:`lars_path` or 
+:func:`lars_path_gram`
 
 Mathematical formulation
 ------------------------

diff --git a/doc/whats_new/v0.21.rst b/doc/whats_new/v0.21.rst
@@ -195,6 +195,11 @@ Support for Python 3.4 and below has been officially dropped.
   :class:`linear_model.LogisticRegressionCV` now support Elastic-Net penalty,
   with the 'saga' solver. :issue:`11646` by :user:`Nicolas Hug <NicolasHug>`.
 
+- |Feature| Added :class:`linear_model.lars_path_gram`, which is
+  :class:`linear_model.lars_path` in the sufficient stats mode, allowing
+  users to compute :class:`linear_model.lars_path` without providing
+  ``X`` and ``y``. :issue:`11699` by :user:`Kuai Yu <yukuairoy>`.
+
 - |Enhancement| :class:`linear_model.make_dataset` now preserves
   ``float32`` and ``float64`` dtypes. :issue:`8769` and :issue:`11000` by
   :user:`Nelle Varoquaux <NelleV>`, :user:`Arthur Imbert <Henley13>`,
@@ -206,6 +211,10 @@ Support for Python 3.4 and below has been officially dropped.
   by the liblinear solver. :issue:`12860` by :user:`Nicolas Hug
   <NicolasHug>`.
 
+- |Enhancement| `sparse_cg` solver in :class:`linear_model.ridge.Ridge`
+  now supports fitting the intercept (i.e. ``fit_intercept=True``) when
+  inputs are sparse. :issue:`13336` by :user:`Bartosz Telenczuk <btel>`.
+
 - |Fix| Fixed a bug in :class:`linear_model.LogisticRegression` and
   :class:`linear_model.LogisticRegressionCV` with 'saga' solver, where the
   weights would not be correctly updated in some cases.
@@ -230,18 +239,18 @@ Support for Python 3.4 and below has been officially dropped.
   ``fit_intercept=True`` in sparse and dense case.
   :issue:`13279` by `Alexandre Gramfort`_
 
-- |API| :func:`linear_model.logistic_regression_path` is deprecated
-  in version 0.21 and will be removed in version 0.23.
-  :issue:`12821` by :user:`Nicolas Hug <NicolasHug>`.
-
-- |Enhancement| `sparse_cg` solver in :class:`linear_model.ridge.Ridge`
-  now supports fitting the intercept (i.e. ``fit_intercept=True``) when
-  inputs are sparse . :issue:`13336` by :user:`Bartosz Telenczuk <btel>`
-
 - |Fix| Fixed a bug in :class:`linear_model.HuberRegressor` that was
   broken when X was of dtype bool.
   :issue:`13328` by `Alexandre Gramfort`_.
 
+- |API|  The use of :class:`linear_model.lars_path` with ``X=None`` 
+  while passing ``Gram`` is deprecated in version 0.21 and will be removed
+  in version 0.23. Use :class:`linear_model.lars_path_gram` instead.
+  :issue:`11699` by :user:`Kuai Yu <yukuairoy>`.
+
+- |API| :func:`linear_model.logistic_regression_path` is deprecated
+  in version 0.21 and will be removed in version 0.23.
+  :issue:`12821` by :user:`Nicolas Hug <NicolasHug>`.
 
 :mod:`sklearn.manifold`
 ............................

diff --git a/sklearn/covariance/graph_lasso_.py b/sklearn/covariance/graph_lasso_.py
@@ -20,8 +20,8 @@
 from ..exceptions import ConvergenceWarning
 from ..utils.validation import check_random_state, check_array
 from ..utils import deprecated
-from ..linear_model import lars_path
 from ..linear_model import cd_fast
+from ..linear_model import lars_path_gram
 from ..model_selection import check_cv, cross_val_score
 from ..utils._joblib import Parallel, delayed
 
@@ -228,8 +228,8 @@ def graphical_lasso(emp_cov, alpha, cov_init=None, mode='cd', tol=1e-4,
                             check_random_state(None), False)
                     else:
                         # Use LARS
-                        _, _, coefs = lars_path(
-                            sub_covariance, row, Xy=row, Gram=sub_covariance,
+                        _, _, coefs = lars_path_gram(
+                            Xy=row, Gram=sub_covariance, n_samples=row.size,
                             alpha_min=alpha / (n_features - 1), copy_Gram=True,
                             eps=eps, method='lars', return_path=False)
                 # Update the precision matrix

diff --git a/sklearn/linear_model/__init__.py b/sklearn/linear_model/__init__.py
@@ -12,8 +12,8 @@
 from .base import LinearRegression
 
 from .bayes import BayesianRidge, ARDRegression
-from .least_angle import (Lars, LassoLars, lars_path, LarsCV, LassoLarsCV,
-                          LassoLarsIC)
+from .least_angle import (Lars, LassoLars, lars_path, lars_path_gram, LarsCV,
+                          LassoLarsCV, LassoLarsIC)
 from .coordinate_descent import (Lasso, ElasticNet, LassoCV, ElasticNetCV,
                                  lasso_path, enet_path, MultiTaskLasso,
                                  MultiTaskElasticNet, MultiTaskElasticNetCV,
@@ -72,6 +72,7 @@
            'TheilSenRegressor',
            'enet_path',
            'lars_path',
+           'lars_path_gram',
            'lasso_path',
            'logistic_regression_path',
            'orthogonal_mp',