Create a base for managing groups of features.

docs/src/api.rst

@@ -15,6 +15,7 @@ Base Classes
 
    base_variable_importance.BaseVariableImportance
    base_perturbation.BasePerturbation
+   base_variable_importance.GroupVariableImportanceMixin
 
 Feature Importance Classes
 ==========================

docs/src/concepts.rst

@@ -3,4 +3,4 @@
 
 ======================
 Definition of concepts
-======================
+======================

docs/src/glm_methods.rst

@@ -9,4 +9,4 @@ GLM methods
    :maxdepth: 2
 
    glm_methods/desparsified_lasso.rst
-   glm_methods/knockoffs.rst
+   glm_methods/knockoffs.rst

docs/src/grouping.rst

@@ -3,4 +3,4 @@
 
 ==========================================
 Measuring the importance of feature groups
-==========================================
+==========================================

docs/src/high_dimension.rst

@@ -3,4 +3,4 @@
 
 ===========================
 Inference in high dimension
-===========================
+===========================

docs/src/marginal_methods.rst

@@ -8,4 +8,4 @@ Marginal methods
 .. toctree::
    :maxdepth: 2
 
-   marginal_methods/leave_one_covariate_in.rst
+   marginal_methods/leave_one_covariate_in.rst

docs/src/model_agnostic_methods/leave_one_covariate_out.rst

@@ -5,4 +5,4 @@
 Leave-One-Covariate-Out
 ========================
 
-TODO: Write this section.
+TODO: Write this section.

docs/src/model_agnostic_methods/total_sobol_index.rst

@@ -16,4 +16,4 @@ where :math:`X^{-j}` denotes the feature vector without the :math:`j^{th}` featu
 :math:`\mu_{-j}(X^{-j})` is the same predictive model as :math:`\mu(X)` but retrained 
 on the reduced feature set :math:`X^{-j}`. When :math:`\mathcal{L}` is the squared loss, 
 for a regression task, :math:`\mu_{-j}(X^{-j}) = \mathbb{E}[y | X^{-j}]` and when 
-:math:`\mathcal{L}` is the log-loss, for a classification task, :math:`\mu_{-j}(X^{-j}) = P(y | X^{-j})`.
+:math:`\mathcal{L}` is the log-loss, for a classification task, :math:`\mu_{-j}(X^{-j}) = P(y | X^{-j})`.

docs/src/visualization.rst

@@ -3,4 +3,4 @@
 
 =======================
 Tools for visualization
-=======================
+=======================

examples/plot_cfi.py

@@ -72,12 +72,14 @@
     loss=log_loss,
     method="predict_proba",
     imputation_model_continuous=RidgeCV(),
+    features_groups={
+        feat_name: [i] for i, feat_name in enumerate(load_wine().feature_names)
+    },
     random_state=0,
 )
 cfi.fit(
     X_train,
     y_train,
-    groups={feat_name: [i] for i, feat_name in enumerate(load_wine().feature_names)},
 )
 importances = cfi.importance(X_test, y_test)
 

examples/plot_importance_classification_iris.py

@@ -69,7 +69,7 @@ def run_one_fold(
     train_index,
     test_index,
     vim_name="CFI",
-    groups=None,
+    features_groups=None,
 ):
     model_c = clone(model)
     model_c.fit(X[train_index], y[train_index])
@@ -94,6 +94,7 @@ def run_one_fold(
             random_state=2,
             method=method,
             loss=loss,
+            features_groups=features_groups,
         )
     elif vim_name == "PFI":
         vim = PFI(
@@ -102,14 +103,15 @@ def run_one_fold(
             random_state=3,
             method=method,
             loss=loss,
+            features_groups=features_groups,
         )
 
-    vim.fit(X[train_index], y[train_index], groups=groups)
+    vim.fit(X[train_index], y[train_index])
     importance = vim.importance(X[test_index], y[test_index])["importance"]
 
     return pd.DataFrame(
         {
-            "feature": groups.keys(),
+            "feature": features_groups.keys(),
             "importance": importance,
             "vim": vim_name,
             "model": model_name,
@@ -140,10 +142,16 @@ def run_one_fold(
     ),
 ]
 cv = KFold(n_splits=5, shuffle=True, random_state=6)
-groups = {ft: [i] for i, ft in enumerate(dataset.feature_names)}
+features_groups = {ft: [i] for i, ft in enumerate(dataset.feature_names)}
 out_list = Parallel(n_jobs=5)(
     delayed(run_one_fold)(
-        X, y, model, train_index, test_index, vim_name=vim_name, groups=groups
+        X,
+        y,
+        model,
+        train_index,
+        test_index,
+        vim_name=vim_name,
+        features_groups=features_groups,
     )
     for train_index, test_index in cv.split(X)
     for model in models
@@ -279,16 +287,22 @@ def plot_results(df_importance, df_pval):
 # mitigate this issue, we can group correlated features together and measure the
 # importance of these feature groups. For instance, we can group 'sepal width' with
 # 'sepal length' and 'petal length' with 'petal width' and the spurious feature.
-groups = {"sepal features": [0, 1], "petal features": [2, 3, 4]}
+features_groups = {"sepal features": [0, 1], "petal features": [2, 3, 4]}
 out_list = Parallel(n_jobs=5)(
     delayed(run_one_fold)(
-        X, y, model, train_index, test_index, vim_name=vim_name, groups=groups
+        X,
+        y,
+        model,
+        train_index,
+        test_index,
+        vim_name=vim_name,
+        features_groups=features_groups,
     )
     for train_index, test_index in cv.split(X)
     for model in models
     for vim_name in ["CFI", "PFI"]
 )
 
-df_grouped = pd.concat(out_list)
-df_pval = compute_pval(df_grouped, threshold=threshold)
-plot_results(df_grouped, df_pval)
+df_features_grouped = pd.concat(out_list)
+df_pval = compute_pval(df_features_grouped, threshold=threshold)
+plot_results(df_features_grouped, df_pval)

pyproject.toml

@@ -58,7 +58,7 @@ style = ["black  >= 24.4.2", "codespell >=2.4.0", "isort  >= 5.13.2"]
 test = [
     "coverage       >= 6.0,     < 8",
     "iniconfig      >= 0.1,     < 3",
-    "matplotlib     >= 3.1.0,   < 4",
+    "matplotlib     >= 3.4.0,   < 4",
     "packaging      >= 14.0,    < 100",
     "pytest         >= 8.0,     < 9",
     "pytest-cov     >= 5.0, < 8",