[ENH] Add LassoCV to nilearn.decoding.DecoderRegressor (#3781)

* add LassoCV to DecoderRegressor

* update decoder tests

* add 'lasso'/'lasso_regressor' to regressor options

* update changelog entry

* use n_alphas=30 as default instead of 100

* add lasso regression decoding to example

* fix formatting in example

* add lasso(_regressor) to list of estimators in docs

* revert example

* fix formatting

doc/changes/latest.rst

@@ -44,6 +44,8 @@ Enhancements
 
 - Make return key names in the description file of destrieux surface consistent with :func:`~datasets.fetch_atlas_surf_destrieux` (:gh:`3774` by `Tarun Samanta`_).
 
+- Add ``LassoCV`` as a new estimator option for Decoder objects (:gh: `3781` by `Michelle Wang`_)
+
 Changes
 -------
 

doc/decoding/estimator_choice.rst

@@ -127,6 +127,8 @@ In :class:`nilearn.decoding.DecoderRegressor` you can use some of these objects
 
 * `ridge_regressor` (same as `ridge`) : `Ridge regression <https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.RidgeCV.html>`_.
 
+* `lasso_regressor` (same as `lasso`) : `Lasso regression <https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LassoCV.html>`_.
+
 * `dummy_regressor` : A `dummy regressor <https://scikit-learn.org/stable/modules/generated/sklearn.dummy.DummyRegressor.html>`_ is a regressor that makes predictions using simple rules. It is useful as a simple baseline to compare with other regressors.
 
 .. note::

nilearn/_utils/docs.py

@@ -634,6 +634,17 @@ def custom_function(vertices):
                 svr = SVR(kernel="linear",
                           max_iter=1e4)
 
+        - `lasso`: \
+            :class:`{Lasso regression} <sklearn.linear_model.LassoCV>`.
+            .. code-block:: python
+
+                lasso = LassoCV()
+
+        - `lasso_regressor`: \
+            :class:`{Lasso regression} <sklearn.linear_model.LassoCV>`.
+            .. note::
+                Same option as `lasso`.
+
         - `dummy_regressor`: \
             :class:`{Dummy regressor} <sklearn.dummy.DummyRegressor>`.
             .. code-block:: python

nilearn/decoding/decoder.py

@@ -24,6 +24,7 @@
 from sklearn import clone
 from sklearn.dummy import DummyClassifier, DummyRegressor
 from sklearn.linear_model import (
+    LassoCV,
     LinearRegression,
     LogisticRegressionCV,
     RidgeClassifierCV,
@@ -59,6 +60,8 @@
     ridge_classifier=RidgeClassifierCV(),
     ridge_regressor=RidgeCV(),
     ridge=RidgeCV(),
+    lasso=LassoCV(),
+    lasso_regressor=LassoCV(),
     svr=SVR(kernel="linear", max_iter=10000),
     dummy_classifier=DummyClassifier(strategy="stratified", random_state=0),
     dummy_regressor=DummyRegressor(strategy="mean"),
@@ -114,6 +117,8 @@ def _check_param_grid(estimator, X, y, param_grid=None):
         elif isinstance(estimator, LogisticRegressionCV):
             param_grid = _replace_param_grid_key(param_grid, "C", "Cs")
             param_grid = _wrap_param_grid(param_grid, "Cs")
+        elif isinstance(estimator, LassoCV):
+            param_grid = _wrap_param_grid(param_grid, "alphas")
 
     return param_grid
 
@@ -155,7 +160,14 @@ def _default_param_grid(estimator, X, y):
             raise NotImplementedError(message)
     elif not isinstance(
         estimator,
-        (LogisticRegressionCV, LinearSVC, RidgeCV, RidgeClassifierCV, SVR),
+        (
+            LogisticRegressionCV,
+            LinearSVC,
+            RidgeCV,
+            RidgeClassifierCV,
+            SVR,
+            LassoCV,
+        ),
     ):
         raise ValueError(
             "Invalid estimator. The supported estimators are:"
@@ -185,6 +197,11 @@ def _default_param_grid(estimator, X, y):
         # so for L2 penalty, param_grid["Cs"] is either 1e-3, ..., 1e4, and
         # for L1 penalty the values are obtained in a more data-driven way
         param_grid["Cs"] = [np.geomspace(2e-3, 2e4, 8) * min_c]
+    elif isinstance(estimator, LassoCV):
+        # the default is to generate 30 alphas based on the data
+        # (alpha values can also be set with the 'alphas' parameter, in which
+        # case 'n_alphas' is ignored)
+        param_grid["n_alphas"] = [30]
     elif isinstance(estimator, (LinearSVC, SVR)):
         # similar logic as above:
         # - for L2 penalty this is [1, 10, 100]
@@ -398,7 +415,7 @@ def _parallel_fit(
                 elif isinstance(estimator, DummyRegressor):
                     dummy_output = estimator.constant_
 
-            if isinstance(estimator, (RidgeCV, RidgeClassifierCV)):
+            if isinstance(estimator, (RidgeCV, RidgeClassifierCV, LassoCV)):
                 params["best_alpha"] = estimator.alpha_
             elif isinstance(estimator, LogisticRegressionCV):
                 params["best_C"] = estimator.C_.item()
@@ -658,8 +675,8 @@ def fit(self, X, y, groups=None):
             built-in cross-validation, this will include an additional key for
             the single best value estimated by the built-in cross-validation
             ('best_C' for LogisticRegressionCV and 'best_alpha' for
-            RidgeCV/RidgeClassifierCV), in addition to the input list of
-            values.
+            RidgeCV/RidgeClassifierCV/LassoCV), in addition to the input list
+            of values.
 
         'scorer_' : function
             Scorer function used on the held out data to choose the best

nilearn/decoding/tests/test_decoder.py

@@ -28,6 +28,7 @@
 from sklearn.ensemble import RandomForestClassifier
 from sklearn.exceptions import NotFittedError
 from sklearn.linear_model import (
+    LassoCV,
     LogisticRegressionCV,
     RidgeClassifierCV,
     RidgeCV,
@@ -140,7 +141,12 @@ def multiclass_data():
 
 
 @pytest.mark.parametrize(
-    "regressor, param", [(RidgeCV(), ["alphas"]), (SVR(kernel="linear"), "C")]
+    "regressor, param",
+    [
+        (RidgeCV(), ["alphas"]),
+        (SVR(kernel="linear"), ["C"]),
+        (LassoCV(), ["n_alphas"]),
+    ],
 )
 def test_check_param_grid_regression(regressor, param):
     """Test several estimators.
@@ -414,6 +420,7 @@ def test_parallel_fit(rand_X_Y):
         (RidgeCV(), "alphas", "best_alpha", False),
         (RidgeClassifierCV(), "alphas", "best_alpha", True),
         (LogisticRegressionCV(), "Cs", "best_C", True),
+        (LassoCV(), "alphas", "best_alpha", False),
     ],
 )
 def test_parallel_fit_builtin_cv(