Mercurial > pylearn

comparison sandbox/rbm/model.py @ 398:6e55ccb7e2bf

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Better output
author Joseph Turian <turian@gmail.com>
date Tue, 08 Jul 2008 20:48:56 -0400
parents e0c9357456e0
children 8796b91a9f09
comparison
equal deleted inserted replaced
397:25a3212287cd 398:6e55ccb7e2bf
28 assert v[j][i] >= 0 and v[j][i] <= 1 28 assert v[j][i] >= 0 and v[j][i] <= 1
29 if random.random() < v[j][i]: x[j][i] = 1 29 if random.random() < v[j][i]: x[j][i] = 1
30 else: x[j][i] = 0 30 else: x[j][i] = 0
31 return x 31 return x
32 32
33 def crossentropy(output, target):
34 """
35 Compute the crossentropy of binary output wrt binary target.
36 @note: We do not sum, crossentropy is computed by component.
37 @todo: Rewrite as a scalar, and then broadcast to tensor.
38 """
39 return -(target * numpy.log(output) + (1 - target) * numpy.log(1 - output))
40
41
33 class Model: 42 class Model:
34 def __init__(self): 43 def __init__(self):
35 self.parameters = parameters.Parameters(randomly_initialize=True) 44 self.parameters = parameters.Parameters(randomly_initialize=True)
36 45
37 def update(self, instance): 46 def update(self, instance):
53 print 62 print
54 print "v[0]:", v0 63 print "v[0]:", v0
55 print "Q(h[0][i] = 1 | v[0]):", q0 64 print "Q(h[0][i] = 1 | v[0]):", q0
56 print "h[0]:", h0 65 print "h[0]:", h0
57 print "P(v[1][j] = 1 | h[0]):", p0 66 print "P(v[1][j] = 1 | h[0]):", p0
67 print "XENT(P(v[1][j] = 1 | h[0]) | v0):", numpy.sum(crossentropy(p0, v0))
58 print "v[1]:", v1 68 print "v[1]:", v1
59 print "Q(h[1][i] = 1 | v[1]):", q1 69 print "Q(h[1][i] = 1 | v[1]):", q1
60 70
61 print self.parameters.w.shape
62 self.parameters.w += LR * (dot(v0.T, h0) - dot(v1.T, q1)) 71 self.parameters.w += LR * (dot(v0.T, h0) - dot(v1.T, q1))
63 self.parameters.b += LR * (h0 - q1) 72 self.parameters.b += LR * (h0 - q1)
64 self.parameters.c += LR * (v0 - v1) 73 self.parameters.c += LR * (v0 - v1)
65 print self.parameters 74 # print self.parameters