From bb9b1f6aa7cd53c9ea4e5a705ee2ecc3fedffea9 Mon Sep 17 00:00:00 2001
From: Nicolas Weber <thakis@67-218-109-209.cust.layer42.net>
Date: Fri, 20 Feb 2009 08:53:58 -0800
Subject: [PATCH] more graphs

---
 drawnumpredict.py | 40 ++++++++++++++++++++++++++++++++++------
 numpredict.py     |  3 ++-
 2 files changed, 36 insertions(+), 7 deletions(-)

diff --git a/drawnumpredict.py b/drawnumpredict.py
index 2dcddcc..8950d6a 100644
--- a/drawnumpredict.py
+++ b/drawnumpredict.py
@@ -2,17 +2,45 @@
 
 from pylab import *
 
+S = 1.0
+
 def cumulativegraph(data, vec1, high, k=5, weightfun=numpredict.gaussianweight):
-  t1 = arange(0.0, high, 1)
+  t1 = arange(0.0, high, S)
   cprob = array([numpredict.probguess(data, vec1, 0, v, k, weightfun)
     for v in t1])
   plot(t1, cprob)
-  show()  # this does user interaction
+
+
+def probabilitygraph(data, vec1, high, k=5, weightfun=numpredict.gaussianweight,
+    ss=5.0/10):
+  t1 = arange(0.0, high, S)
+  probs = [numpredict.probguess(data, vec1, v, v+S, k, weightfun) for v in t1]
+
+  # gaussian smooth with nearby points
+  smoothed = []
+  for i in range(len(probs)):  # O(n^2)
+    sv = 0.0
+    for j in range(0, len(probs)):
+      dist = abs(i - j)*0.1
+      weight = numpredict.gaussianweight(dist, sigma=ss)
+      sv += weight*probs[j]
+    smoothed.append(sv)
+
+  plot(t1, array(probs))
+  plot(t1, array(smoothed))
+
 
 if __name__ == '__main__':
   s = numpredict.wineset3(k=500)
 
-  # Draw graph that shows p(price | [rating=99, age=20])
-  print 'Real price 1:', numpredict.wineprice(99.0, 20.0)
-  print 'Real price 2:', 0.6 * numpredict.wineprice(99.0, 20.0)
-  cumulativegraph(s, [99.0, 20.0], 120)
+  import pprint
+  pprint.pprint(s)
+
+  #wine = [99.0, 20.0]  # choose params were the sample data is somewhat dense
+  wine = [99.0, 20.0]  # choose params were the sample data is somewhat dense
+  # Draw graph that shows p(price | wine)
+  print 'Real price 1:', numpredict.wineprice(wine[0], wine[1])
+  print 'Real price 2:', 0.6 * numpredict.wineprice(wine[0], wine[1])
+  cumulativegraph(s, wine, 120)
+  probabilitygraph(s, wine, 120)
+  show()  # this does user interaction
diff --git a/numpredict.py b/numpredict.py
index 654968c..678368b 100644
--- a/numpredict.py
+++ b/numpredict.py
@@ -71,7 +71,8 @@ def subtractweight(dist, const=1.0):
   return max(0, const - dist)
 
 
-def gaussianweight(dist, sigma=10.0):
+#def gaussianweight(dist, sigma=10.0):
+def gaussianweight(dist, sigma=5.0):
   return math.exp(-0.5 * (dist/sigma)**2)