DOC add more details in the examples

Author: TomDLT

tutorials/movies_3T/01_plot_explainable_variance.py

@@ -14,7 +14,8 @@
 This shared component can be estimated by taking the mean over repeats of the
 same experiment. The variance of this shared component, that we call the
 explainable variance, is the upper bound of the voxelwise modeling
-performances.
+performances. The explainable variance is also sometimes called the *noise
+ceiling*.
 """
 # sphinx_gallery_thumbnail_number = 2
 ###############################################################################

tutorials/movies_3T/02_plot_wordnet_model.py

@@ -120,6 +120,15 @@
 # mean value in fMRI recording is non-informative, so each run is detrended and
 # demeaned independently, and we do not need to predict an intercept value in
 # the linear model.
+#
+# However, we prefer not to normalize by the standard deviation of each
+# feature. Indeed, if the features are extracted in a consistent way from the
+# stimulus, there relative scale is meaningful. Normalizing them independently
+# from each other would remove this meaning. Moreover, the wordnet features are
+# one-hot-encoded, which means that each feature is either present (1) or not
+# present (0) in each sample. Normalizing one-hot-encoded features is not
+# recommended, since it would scale disproportionately the infrequent features.
+
 from sklearn.preprocessing import StandardScaler
 scaler = StandardScaler(with_mean=True, with_std=False)
 
@@ -398,6 +407,8 @@
 plt.show()
 
 ###############################################################################
+# In this dataset, the brain responses are recorded every two seconds.
+#
 # We see that the hemodynamic response function (HRF) is captured in the model
 # weights. In practice, we can limit the number of features by using only
 # the most informative delays, for example [1, 2, 3, 4].