[DOC] Speed up fit computation with parallel processing (#4108)

* Speed up compute with n_jobs * Set n_jobs to max cpu where possible * [full doc] test full build * Set n_jobs to 2 * [full doc] test full build njobs=2 * Update examples/02_decoding/plot_haxby_searchlight.py * Add the rest
nilearn · Nov 20, 2023 · f4b03d4 · f4b03d4
1 parent b8053f1
commit f4b03d4
Show file tree

Hide file tree

Showing 24 changed files with 38 additions and 26 deletions.
diff --git a/examples/02_decoding/plot_haxby_frem.py b/examples/02_decoding/plot_haxby_frem.py
@@ -55,7 +55,7 @@
 # --------
 from nilearn.decoding import FREMClassifier
 
-decoder = FREMClassifier(cv=10, standardize="zscore_sample")
+decoder = FREMClassifier(cv=10, standardize="zscore_sample", n_jobs=2)
 # Fit model on train data and predict on test data
 decoder.fit(X_train, y_train)
 y_pred = decoder.predict(X_test)

diff --git a/examples/02_decoding/plot_haxby_searchlight.py b/examples/02_decoding/plot_haxby_searchlight.py
@@ -67,7 +67,7 @@
 # Make processing parallel
 # /!\ As each thread will print its progress, n_jobs > 1 could mess up the
 #     information output.
-n_jobs = 1
+n_jobs = 2
 
 # Define the cross-validation scheme used for validation.
 # Here we use a KFold cross-validation on the session, which corresponds to

diff --git a/examples/02_decoding/plot_haxby_searchlight_surface.py b/examples/02_decoding/plot_haxby_searchlight_surface.py
@@ -78,8 +78,7 @@
 cv = KFold(n_splits=3, shuffle=False)
 
 # Cross-validated search light
-scores = search_light(X, y, estimator, adjacency, cv=cv, n_jobs=1)
-
+scores = search_light(X, y, estimator, adjacency, cv=cv, n_jobs=2)
 # %%
 # Visualization
 # -------------

diff --git a/examples/02_decoding/plot_mixed_gambles_frem.py b/examples/02_decoding/plot_mixed_gambles_frem.py
@@ -80,6 +80,7 @@
     penalty="tv-l1",
     eps=1e-1,  # prefer large alphas
     memory="nilearn_cache",
+    n_jobs=2,
 )
 # tv_l1.fit(zmap_filenames, behavioral_target)
 # plot_stat_map(tv_l1.coef_img_, title="TV-L1", display_mode="yz",

diff --git a/examples/02_decoding/plot_miyawaki_encoding.py b/examples/02_decoding/plot_miyawaki_encoding.py
@@ -60,7 +60,10 @@
 from nilearn.maskers import MultiNiftiMasker
 
 masker = MultiNiftiMasker(
-    mask_img=dataset.mask, detrend=True, standardize="zscore_sample"
+    mask_img=dataset.mask,
+    detrend=True,
+    standardize="zscore_sample",
+    n_jobs=2,
 )
 masker.fit()
 fmri_data = masker.transform(fmri_random_runs_filenames)

diff --git a/examples/02_decoding/plot_miyawaki_reconstruction.py b/examples/02_decoding/plot_miyawaki_reconstruction.py
@@ -63,7 +63,7 @@
 
 # Load and mask fMRI data
 masker = MultiNiftiMasker(
-    mask_img=miyawaki_dataset.mask, detrend=True, standardize=False
+    mask_img=miyawaki_dataset.mask, detrend=True, standardize=False, n_jobs=2
 )
 masker.fit()
 X_train = masker.transform(X_random_filenames)

diff --git a/examples/02_decoding/plot_oasis_vbm.py b/examples/02_decoding/plot_oasis_vbm.py
@@ -128,7 +128,7 @@
     mask=mask,
     scoring="neg_mean_absolute_error",
     screening_percentile=1,
-    n_jobs=1,
+    n_jobs=2,
     standardize="zscore_sample",
 )
 # Fit and predict with the decoder
@@ -197,7 +197,7 @@
     data,  # + intercept as a covariate by default
     n_perm=2000,  # 1,000 in the interest of time; 10000 would be better
     verbose=1,  # display progress bar
-    n_jobs=1,
+    n_jobs=2,
 )
 signed_neg_log_pvals = neg_log_pvals * np.sign(t_scores_original_data)
 signed_neg_log_pvals_unmasked = nifti_masker.inverse_transform(

diff --git a/examples/02_decoding/plot_oasis_vbm_space_net.py b/examples/02_decoding/plot_oasis_vbm_space_net.py
@@ -60,6 +60,7 @@
     screening_percentile=5.0,
     memory_level=2,
     standardize="zscore_sample",
+    n_jobs=2,
 )
 decoder.fit(gm_imgs_train, age_train)  # fit
 coef_img = decoder.coef_img_

diff --git a/examples/02_decoding/plot_simulated_data.py b/examples/02_decoding/plot_simulated_data.py
@@ -183,7 +183,7 @@ def plot_slices(data, title=None):
             estimator=svm.SVR(kernel="linear"),
             cv=KFold(n_splits=4),
             verbose=1,
-            n_jobs=1,
+            n_jobs=2,
         ),
     ),
 ]

diff --git a/examples/03_connectivity/plot_compare_decomposition.py b/examples/03_connectivity/plot_compare_decomposition.py
@@ -54,6 +54,7 @@
     mask_strategy="whole-brain-template",
     random_state=0,
     standardize="zscore_sample",
+    n_jobs=2,
 )
 canica.fit(func_filenames)
 
@@ -116,6 +117,7 @@
     n_epochs=1,
     mask_strategy="whole-brain-template",
     standardize="zscore_sample",
+    n_jobs=2,
 )
 
 print("[Example] Fitting dictionary learning model")

diff --git a/examples/04_glm_first_level/plot_bids_features.py b/examples/04_glm_first_level/plot_bids_features.py
@@ -90,6 +90,7 @@
     space_label,
     smoothing_fwhm=5.0,
     derivatives_folder=derivatives_folder,
+    n_jobs=2,
 )
 
 # %%

diff --git a/examples/05_glm_second_level/plot_oasis.py b/examples/05_glm_second_level/plot_oasis.py
@@ -103,7 +103,9 @@
 # also smooth a little bit to improve statistical behavior.
 from nilearn.glm.second_level import SecondLevelModel
 
-second_level_model = SecondLevelModel(smoothing_fwhm=2.0, mask_img=mask_img)
+second_level_model = SecondLevelModel(
+    smoothing_fwhm=2.0, mask_img=mask_img, n_jobs=2
+)
 second_level_model.fit(
     gray_matter_map_filenames,
     design_matrix=design_matrix,

diff --git a/examples/05_glm_second_level/plot_proportion_activated_voxels.py b/examples/05_glm_second_level/plot_proportion_activated_voxels.py
@@ -45,7 +45,7 @@
 # Model specification and fit
 from nilearn.glm.second_level import SecondLevelModel
 
-second_level_model = SecondLevelModel(smoothing_fwhm=8.0)
+second_level_model = SecondLevelModel(smoothing_fwhm=8.0, n_jobs=2)
 second_level_model = second_level_model.fit(
     second_level_input, design_matrix=design_matrix
 )

diff --git a/examples/05_glm_second_level/plot_second_level_association_test.py b/examples/05_glm_second_level/plot_second_level_association_test.py
@@ -68,7 +68,7 @@
 # Fit of the second-level model
 from nilearn.glm.second_level import SecondLevelModel
 
-model = SecondLevelModel(smoothing_fwhm=5.0)
+model = SecondLevelModel(smoothing_fwhm=5.0, n_jobs=2)
 model.fit(contrast_map_filenames, design_matrix=design_matrix)
 
 # %%
@@ -143,7 +143,7 @@
     two_sided_test=False,
     mask=None,
     smoothing_fwhm=5.0,
-    n_jobs=1,
+    n_jobs=2,
 )
 
 # %%

diff --git a/examples/05_glm_second_level/plot_second_level_one_sample_test.py b/examples/05_glm_second_level/plot_second_level_one_sample_test.py
@@ -80,7 +80,7 @@
 # Next, we specify the model and fit it.
 from nilearn.glm.second_level import SecondLevelModel
 
-second_level_model = SecondLevelModel(smoothing_fwhm=8.0)
+second_level_model = SecondLevelModel(smoothing_fwhm=8.0, n_jobs=2)
 second_level_model = second_level_model.fit(
     second_level_input,
     design_matrix=design_matrix,
@@ -167,7 +167,7 @@
     n_perm=500,  # 500 for the sake of time. Ideally, this should be 10,000.
     two_sided_test=False,
     smoothing_fwhm=8.0,
-    n_jobs=1,
+    n_jobs=2,
     threshold=0.001,
 )
 

diff --git a/examples/05_glm_second_level/plot_second_level_two_sample_test.py b/examples/05_glm_second_level/plot_second_level_two_sample_test.py
@@ -100,11 +100,11 @@
 # We specify the analysis models and fit them.
 from nilearn.glm.second_level import SecondLevelModel
 
-second_level_model_unpaired = SecondLevelModel().fit(
+second_level_model_unpaired = SecondLevelModel(n_jobs=2).fit(
     second_level_input, design_matrix=unpaired_design_matrix
 )
 
-second_level_model_paired = SecondLevelModel().fit(
+second_level_model_paired = SecondLevelModel(n_jobs=2).fit(
     second_level_input, design_matrix=paired_design_matrix
 )
 

diff --git a/examples/05_glm_second_level/plot_thresholding.py b/examples/05_glm_second_level/plot_thresholding.py
@@ -40,7 +40,7 @@
 # Next, we specify and estimate the model.
 from nilearn.glm.second_level import SecondLevelModel
 
-second_level_model = SecondLevelModel().fit(
+second_level_model = SecondLevelModel(n_jobs=2).fit(
     cmap_filenames, design_matrix=design_matrix
 )
 

diff --git a/examples/07_advanced/plot_advanced_decoding_scikit.py b/examples/07_advanced/plot_advanced_decoding_scikit.py
@@ -126,7 +126,7 @@
     cv=cv,
     scoring="roc_auc",
     groups=session_label,
-    n_jobs=-1,
+    n_jobs=2,
 )
 print(f"SVC accuracy (tuned parameters): {cv_scores.mean():.3f}")
 
@@ -274,6 +274,6 @@
 #                             fmri_masked,
 #                             target,
 #                             cv=cv,
-#                             n_jobs=-1,
+#                             n_jobs=2,
 #                             verbose=1)
 # But, be aware that this can take * A WHILE * ...
diff --git a/examples/07_advanced/plot_age_group_prediction_cross_val.py b/examples/07_advanced/plot_age_group_prediction_cross_val.py
@@ -107,7 +107,7 @@
     cv=cv,
     verbose=1,
     refit=False,
-    n_jobs=8,
+    n_jobs=2,
 )
 gs.fit(masked_data, classes)
 mean_scores = gs.cv_results_["mean_test_score"]

diff --git a/examples/07_advanced/plot_beta_series.py b/examples/07_advanced/plot_beta_series.py
@@ -79,6 +79,7 @@
     data_dir,
     "languagelocalizer",
     img_filters=[("desc", "preproc")],
+    n_jobs=2,
 )
 
 # Grab the first subject's model, functional file, and events DataFrame

diff --git a/examples/07_advanced/plot_bids_analysis.py b/examples/07_advanced/plot_bids_analysis.py
@@ -57,7 +57,7 @@
     models_events,
     models_confounds,
 ) = first_level_from_bids(
-    data_dir, task_label, img_filters=[("desc", "preproc")]
+    data_dir, task_label, img_filters=[("desc", "preproc")], n_jobs=2
 )
 
 # %%
@@ -137,7 +137,7 @@
 
 # %%
 # Note that we apply a smoothing of 8mm.
-second_level_model = SecondLevelModel(smoothing_fwhm=8.0)
+second_level_model = SecondLevelModel(smoothing_fwhm=8.0, n_jobs=2)
 second_level_model = second_level_model.fit(second_level_input)
 
 # %%

diff --git a/examples/07_advanced/plot_haxby_mass_univariate.py b/examples/07_advanced/plot_haxby_mass_univariate.py
@@ -117,7 +117,7 @@
     n_perm=10000,
     two_sided_test=True,
     verbose=1,  # display progress bar
-    n_jobs=1,  # can be changed to use more CPUs
+    n_jobs=2,  # can be changed to use more CPUs
 )
 signed_neg_log_pvals = neg_log_pvals * np.sign(t_scores_original_data)
 signed_neg_log_pvals_unmasked = nifti_masker.inverse_transform(

diff --git a/examples/07_advanced/plot_localizer_mass_univariate_methods.py b/examples/07_advanced/plot_localizer_mass_univariate_methods.py
@@ -99,7 +99,7 @@
     tfce=True,
     n_perm=200,  # 200 for the sake of time. Ideally, this should be 10000.
     verbose=1,  # display progress bar
-    n_jobs=1,  # can be changed to use more CPUs
+    n_jobs=2,  # can be changed to use more CPUs
     output_type="dict",
 )
 neg_log_pvals_permuted_ols_unmasked = nifti_masker.inverse_transform(

diff --git a/examples/07_advanced/plot_surface_bids_analysis.py b/examples/07_advanced/plot_surface_bids_analysis.py
@@ -53,7 +53,9 @@
 _, models_run_imgs, models_events, models_confounds = \
     first_level_from_bids(
         data_dir, task_label,
-        img_filters=[('desc', 'preproc')])
+        img_filters=[('desc', 'preproc')],
+        n_jobs=2
+    )
 
 # %%
 # We also need to get the :term:`TR` information.