From: Francois Fleuret <francois@fleuret.org>
Date: Wed, 18 Dec 2019 15:51:25 +0000 (+0100)
Subject: Update.
X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=commitdiff_plain;h=b0e213a468726f721990fdbe0837cc6cdce10427;p=pytorch.git

Update.
---

diff --git a/mi_estimator.py b/mi_estimator.py
index f8b859d..02e9db9 100755
--- a/mi_estimator.py
+++ b/mi_estimator.py
@@ -13,7 +13,8 @@
 # You should have received a copy of the GNU General Public License     #
 # along with this program. If not, see <http://www.gnu.org/licenses/>.  #
 #                                                                       #
-# Written by and Copyright (C) Francois Fleuret                         #
+# Written by Francois Fleuret, (C) Idiap Research Institute             #
+#                                                                       #
 # Contact <francois.fleuret@idiap.ch> for comments & bug reports        #
 #########################################################################
 
@@ -38,19 +39,20 @@ else:
 parser = argparse.ArgumentParser(
     description = '''An implementation of a Mutual Information estimator with a deep model
 
-Three different toy data-sets are implemented:
+    Three different toy data-sets are implemented, each consists of
+    pairs of samples, that may be from different spaces:
 
- (1) Two MNIST images of same class. The "true" MI is the log of the
-     number of used MNIST classes.
+    (1) Two MNIST images of same class. The "true" MI is the log of the
+    number of used MNIST classes.
 
- (2) One MNIST image and a pair of real numbers whose difference is
-     the class of the image. The "true" MI is the log of the number of
-     used MNIST classes.
+    (2) One MNIST image and a pair of real numbers whose difference is
+    the class of the image. The "true" MI is the log of the number of
+    used MNIST classes.
 
- (3) Two 1d sequences, the first with a single peak, the second with
-     two peaks, and the height of the peak in the first is the
-     difference of timing of the peaks in the second. The "true" MI is
-     the log of the number of possible peak heights.''',
+    (3) Two 1d sequences, the first with a single peak, the second with
+    two peaks, and the height of the peak in the first is the
+    difference of timing of the peaks in the second. The "true" MI is
+    the log of the number of possible peak heights.''',
 
     formatter_class = argparse.ArgumentDefaultsHelpFormatter
 )
@@ -197,6 +199,8 @@ def create_image_values_pairs(train = False):
 
 ######################################################################
 
+#
+
 def create_sequences_pairs(train = False):
     nb, length = 10000, 1024
     noise_level = 2e-2
@@ -229,9 +233,6 @@ def create_sequences_pairs(train = False):
     noise = b.new(b.size()).normal_(0, noise_level)
     b = b + noise
 
-    # a = (a - a.mean()) / a.std()
-    # b = (b - b.mean()) / b.std()
-
     return a, b, ha
 
 ######################################################################