[Scipy-svn] r3140 - in trunk/Lib/sandbox/pyem: doc examples

[scipy-svn at scipy.org]

Author: cdavid
Date: 2007-07-02 07:58:40 -0500 (Mon, 02 Jul 2007)
New Revision: 3140

Modified:
   trunk/Lib/sandbox/pyem/doc/Makefile
   trunk/Lib/sandbox/pyem/doc/index.txt
   trunk/Lib/sandbox/pyem/doc/tutorial.pdf
   trunk/Lib/sandbox/pyem/examples/discriminant_analysis.py
Log:
Update doc

Modified: trunk/Lib/sandbox/pyem/doc/Makefile
===================================================================
--- trunk/Lib/sandbox/pyem/doc/Makefile	2007-07-02 12:22:31 UTC (rev 3139)
+++ trunk/Lib/sandbox/pyem/doc/Makefile	2007-07-02 12:58:40 UTC (rev 3140)
@@ -1,4 +1,4 @@
-# Last Change: Sat Jun 09 05:00 PM 2007 J
+# Last Change: Mon Jul 02 09:00 PM 2007 J
 
 # This makefile is used to build the pdf from the rest file and inlined code
 # from python examples
@@ -7,7 +7,7 @@
 rst2tex	= PYTHONPATH=/home/david/local/lib/python2.4/site-packages rst2newlatex.py \
 		  --stylesheet-path base.tex --user-stylesheet user.tex 
 
-pytexfiles	= pyem.tex basic_example1.tex basic_example2.tex basic_example3.tex pdfestimation.tex
+pytexfiles	= pyem.tex basic_example1.tex basic_example2.tex basic_example3.tex pdfestimation.tex discriminant_analysis.tex
 
 SOURCEPATH	= $(PWD)
 
@@ -36,6 +36,9 @@
 pdfestimation.tex: ../examples/pdfestimation.py
 	$(py2tex) $< > $@
 
+discriminant_analysis.tex: ../examples/discriminant_analysis.py
+	$(py2tex) $< > $@
+
 clean:
 	for i in $(pytexfiles); do \
 		rm -f `echo $$i`; \

Modified: trunk/Lib/sandbox/pyem/doc/index.txt
===================================================================
--- trunk/Lib/sandbox/pyem/doc/index.txt	2007-07-02 12:22:31 UTC (rev 3139)
+++ trunk/Lib/sandbox/pyem/doc/index.txt	2007-07-02 12:58:40 UTC (rev 3140)
@@ -13,7 +13,7 @@
         file: Bic_example.png
     /restindex
 
-.. Last Change: Sat Jun 09 07:00 PM 2007 J
+.. Last Change: Mon Jul 02 09:00 PM 2007 J
 
 ===================================================
  PyEM, a python package for Gaussian mixture models
@@ -204,8 +204,26 @@
 Using PyEM for supervised learning
 ----------------------------------
 
-TODO
+The following example shows how to do classification using discriminative
+analysis on the famous iris dataset. The idea is to model each class
+distribution P(data|class) by a mixture of Gaussian, and to classify the data
+with Maximum A Posteriori (eg takes the class which maximizes P(class | data)).
+Not surprisingly, the errors lie on the border between the two classes which
+are not linearly separable.
 
+.. raw:: latex
+
+    \input{discriminant_analysis.tex}
+
+.. figure:: dclass.png 
+    :width: 600
+    :height: 400
+
+    learning data points on the left with isodensity curves as estimated by
+    each mixture. On the right, the misclassified points are in red, green are
+    correctly classified.
+
+
 Note on performances
 ====================
 

Modified: trunk/Lib/sandbox/pyem/doc/tutorial.pdf
===================================================================
(Binary files differ)

Modified: trunk/Lib/sandbox/pyem/examples/discriminant_analysis.py
===================================================================
--- trunk/Lib/sandbox/pyem/examples/discriminant_analysis.py	2007-07-02 12:22:31 UTC (rev 3139)
+++ trunk/Lib/sandbox/pyem/examples/discriminant_analysis.py	2007-07-02 12:58:40 UTC (rev 3140)
@@ -11,6 +11,7 @@
 
 import numpy as N
 import pylab as P
+import matplotlib as MPL
 
 from scipy.sandbox import pyem
 import utils
@@ -93,8 +94,11 @@
 # Plot the results
 #-----------------
 csym = {'setosa' : 's', 'versicolor' : 'x', 'virginica' : 'o'}
-P.figure()
+r = 50.
+P.figure(figsize = [600/r, 400/r])
 
+prop = MPL.font_manager.FontProperties(size='smaller')
+
 # Plot the learning data with the mixtures
 P.subplot(2, 1, 1)
 for i in lmod.values():
@@ -104,6 +108,8 @@
 
 for i in cnames:
     P.plot(ldata[i][:, 0], ldata[i][:, 1], csym[i], label = i + ' (learning)')
+P.xlabel('sepal width')
+P.ylabel('petal width')
 P.legend(loc = 'best')
 
 # Plot the results on test dataset (green for correctly classified, red for
@@ -115,5 +121,7 @@
     if len(incorrect[i]) > 0:
         P.plot(tx[incorrect[i]], ty[incorrect[i]], 'r' + csym[i], 
                label = '%s (incorrectly classified)' % i)
-P.legend(loc = 'best')
-P.show()
+P.legend(loc = 'best', prop = prop)
+P.xlabel('sepal width')
+P.ylabel('petal width')
+P.savefig('dclass.png', dpi = 60)