Reformat the gaussian conditional sampler (#366)

* gaussian distribution reformat

* fix knockoff

* move conditional sampler in statistical tools

* move test from knockoff

* move test in right folder

* remove gaussian

* remove unsed import

* fix center to knockoff

* Change name of GaussianDistribution

* remove parameter assumed_centerd of tests

* change the way of using the randomgenerator

* Apply suggestions from code review

Co-authored-by: Joseph Paillard <[email protected]>

* reformat docstring

* Update test/statistical_tools/test_gaussian_knockoffs.py

Co-authored-by: Joseph Paillard <[email protected]>

* improve tests

* Apply suggestions from code review

Co-authored-by: bthirion <[email protected]>

* add assert

* fix tests

* Improve test

* remove not necessary test

* Update the conditioal sampler for new fit signature

* fix test

* fix isort

* fix examples

* fix documentation

* fix documentation

* update the user guide.

* try to fix the end of file

* Apply suggestion from @bthirion

Co-authored-by: bthirion <[email protected]>

* update makefile

* Add comment

[skip tests]

* update command

* [skip tests]

* fix makefile

* [skip tests]

* change signature of sample

* remove fix

* remove documentation

* Update src/hidimstat/statistical_tools/gaussian_knockoffs.py

Co-authored-by: Joseph Paillard <[email protected]>

* Update src/hidimstat/statistical_tools/gaussian_knockoffs.py

Co-authored-by: Joseph Paillard <[email protected]>

* Update src/hidimstat/statistical_tools/gaussian_knockoffs.py

Co-authored-by: Joseph Paillard <[email protected]>

* Update src/hidimstat/statistical_tools/gaussian_knockoffs.py

Co-authored-by: Joseph Paillard <[email protected]>

* format file

* add test with repeat

---------

Co-authored-by: Joseph Paillard <[email protected]>
Co-authored-by: bthirion <[email protected]>

Author: 3 people

docs/src/api.rst

@@ -52,7 +52,8 @@ Samplers
    :toctree: ./generated/api/class/
    :template: class.rst
 
-   conditional_sampling.ConditionalSampler
+   statistical_tools.conditional_sampling.ConditionalSampler
+   statistical_tools.gaussian_knockoffs.GaussianKnockoffs
 
 Helper Functions
 ================

docs/src/user_guide.rst

@@ -21,4 +21,3 @@ Table of contents
    visualization.rst
    grouping.rst
    high_dimension.rst
-   

examples/plot_pitfalls_permutation_importance.py

@@ -267,7 +267,7 @@
 
 from matplotlib.lines import Line2D
 
-from hidimstat.conditional_sampling import ConditionalSampler
+from hidimstat.statistical_tools.conditional_sampling import ConditionalSampler
 
 X_train, X_test = train_test_split(
     X,

src/hidimstat/conditional_feature_importance.py

@@ -4,7 +4,7 @@
 from sklearn.metrics import root_mean_squared_error
 
 from hidimstat.base_perturbation import BasePerturbation
-from hidimstat.conditional_sampling import ConditionalSampler
+from hidimstat.statistical_tools.conditional_sampling import ConditionalSampler
 
 
 class CFI(BasePerturbation):

src/hidimstat/gaussian_knockoff.py

@@ -7,12 +7,8 @@
 from sklearn.preprocessing import StandardScaler
 from sklearn.utils.validation import check_memory
 
-from hidimstat._utils.utils import check_random_state
-from hidimstat.gaussian_knockoff import (
-    gaussian_knockoff_generation,
-    repeat_gaussian_knockoff_generation,
-)
 from hidimstat.statistical_tools.aggregation import quantile_aggregation
+from hidimstat.statistical_tools.gaussian_knockoffs import GaussianKnockoffs
 from hidimstat.statistical_tools.multiple_testing import fdr_threshold
 
 
@@ -188,36 +184,15 @@ def model_x_knockoff(
     n_jobs = min(n_bootstraps, n_jobs)
     parallel = Parallel(n_jobs, verbose=joblib_verbose)
 
-    # get the seed for the different run
-    rng = check_random_state(random_state)
-    children_rng = rng.spawn(n_bootstraps)
-
     if centered:
         X = StandardScaler().fit_transform(X)
 
-    # estimation of X distribution
-    # original implementation:
-    # https://github.com/msesia/knockoff-filter/blob/master/R/knockoff/R/create_second_order.R
-    mu = X.mean(axis=0)
-    sigma = cov_estimator.fit(X).covariance_
-
     # Create knockoff variables
-    X_tilde, mu_tilde, sigma_tilde_decompose = memory.cache(
-        gaussian_knockoff_generation
-    )(X, mu, sigma, random_state=children_rng[0], tol=tol_gauss)
-
-    if n_bootstraps == 1:
-        X_tildes = [X_tilde]
-    else:
-        X_tildes = parallel(
-            delayed(repeat_gaussian_knockoff_generation)(
-                mu_tilde,
-                sigma_tilde_decompose,
-                random_state=seed,
-            )
-            for seed in children_rng[1:]
-        )
-        X_tildes.insert(0, X_tilde)
+    conditionnal_sampler = GaussianKnockoffs(cov_estimator, tol=tol_gauss)
+    conditionnal_sampler.fit(X)
+    X_tildes = conditionnal_sampler.sample(
+        n_repeats=n_bootstraps, random_state=random_state
+    )
 
     results = parallel(
         delayed(memory.cache(_stat_coefficient_diff))(