Refactored nilearn.decomposition + dict_learning

examples/connectivity/plot_dict_learning_resting_state.py

@@ -0,0 +1,71 @@
+"""
+Group analysis of resting-state fMRI with dictionary learning: DictLearning
+=====================================================
+
+An example applying dictionary learning to resting-state data. This example applies it
+to 10 subjects of the ADHD200 datasets.
+
+Dictionary learning is a sparsity based decomposition method for extracting spatial maps.
+
+    * Gael Varoquaux et al.
+    Multi-subject dictionary learning to segment an atlas of brain spontaneous activity
+    Information Processing in Medical Imaging, 2011, pp. 562-573, Lecture Notes in Computer Science
+
+Pre-prints for paper is available on hal
+(http://hal.archives-ouvertes.fr)
+"""
+from joblib import Memory
+
+### Load ADHD rest dataset ####################################################
+from nilearn import datasets
+# For linear assignment (should be moved in non user space...)
+
+adhd_dataset = datasets.fetch_adhd(n_subjects=20)
+func_filenames = adhd_dataset.func  # list of 4D nifti files for each subject
+
+# print basic information on the dataset
+print('First functional nifti image (4D) is at: %s' %
+      adhd_dataset.func[0])  # 4D data
+
+### Apply DictLearning ########################################################
+from nilearn.decomposition import DictLearning, CanICA
+
+n_components = 10
+
+dict_learning = DictLearning(n_components=n_components, smoothing_fwhm=6.,
+                             memory="nilearn_cache", memory_level=5, verbose=2, random_state=0,
+                             n_jobs=1, alpha=6, n_iter=1000)
+canica = CanICA(n_components=n_components, smoothing_fwhm=6., memory_level=5, verbose=2, random_state=0,
+                memory="nilearn_cache", n_jobs=1, n_init=1, threshold=3.)
+
+estimators = [canica, dict_learning]
+
+components_imgs = []
+
+for estimator in estimators:
+    print('[Example] Learning maps using %s model' % type(estimator).__name__)
+    estimator.fit(func_filenames)
+    print('[Example] Dumping results')
+    components_img = estimator.masker_.inverse_transform(estimator.components_)
+    components_img.to_filename('%s_resting_state.nii.gz' % type(estimator).__name__)
+    components_imgs.append(components_img)
+
+### Visualize the results #####################################################
+# Show some interesting components
+import matplotlib.pyplot as plt
+from nilearn.plotting import plot_stat_map, find_xyz_cut_coords
+from nilearn.image import index_img
+
+mem = Memory(cachedir='~/nilearn_cache')
+
+print('[Example] Displaying')
+
+for i in range(n_components):
+    if i % 2 == 0:
+        fig, axes = plt.subplots(nrows=2)
+        cut_coords = find_xyz_cut_coords(index_img(components_imgs[1], i))
+        for estimator, cur_img, ax in zip(estimators, components_imgs, axes):
+            plot_stat_map(index_img(cur_img, i), title="Component %d" % i, axes=ax,
+                          cut_coords=cut_coords, colorbar=False)
+
+plt.show()

nilearn/decomposition/__init__.py

@@ -3,5 +3,6 @@
 variant of the ICA called Canonnical ICA.
 """
 from .canica import CanICA
+from .dict_learning import DictLearning
 
-__all__ = ['CanICA']
+__all__ = ['CanICA', 'DictLearning']