Mercurial > pylearn

comparison sparse_random_autoassociator/model.py @ 372:75bab24bb2d8

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Moved more logic into model.py
author Joseph Turian <turian@gmail.com>
date Mon, 07 Jul 2008 02:06:15 -0400
parents a1bbcde6b456
children 42cc94cf6c12
comparison
equal deleted inserted replaced
371:22463a194c90 372:75bab24bb2d8
3 Weston's sampling trick (2008). 3 Weston's sampling trick (2008).
4 """ 4 """
5 5
6 from graph import trainfn 6 from graph import trainfn
7 import parameters 7 import parameters
8
9 import globals
10 from globals import LR
11
8 import numpy 12 import numpy
9 from globals import LR 13 import random
14 random.seed(globals.SEED)
15
16 def _select_indices(instance):
17 """
18 Choose nonzero and zero indices (feature columns) of the instance.
19 We select B{all} nonzero indices.
20 We select L{globals.ZERO_SAMPLE_SIZE} zero indices randomly,
21 without replacement.
22 @bug: If there are not ZERO_SAMPLE_SIZE zeroes, we will enter
23 an endless loop.
24 @return: (nonzero_indices, zero_indices)
25 """
26 # Get the nonzero indices
27 nonzero_indices = instance.keys()
28 nonzero_indices.sort()
29
30 # Get the zero indices
31 # @bug: If there are not ZERO_SAMPLE_SIZE zeroes, we will enter an endless loop.
32 zero_indices = []
33 while len(zero_indices) < globals.ZERO_SAMPLE_SIZE:
34 idx = random.randint(0, globals.INPUT_DIMENSION - 1)
35 if idx in nonzero_indices or idx in zero_indices: continue
36 zero_indices.append(idx)
37 zero_indices.sort()
38
39 return (nonzero_indices, zero_indices)
10 40
11 class Model: 41 class Model:
12 def __init__(self): 42 def __init__(self):
13 self.parameters = parameters.Parameters(randomly_initialize=True) 43 self.parameters = parameters.Parameters(randomly_initialize=True)
14 44
15 def update(self, instance, nonzero_indexes, zero_indexes): 45 def update(self, instance):
16 xnonzero = numpy.asarray([instance[idx] for idx in nonzero_indexes]) 46 """
47 Update the L{Model} using one training instance.
48 @param instance: A dict from feature index to (non-zero) value.
49 @todo: Should assert that nonzero_indices and zero_indices
50 are correct (i.e. are truly nonzero/zero).
51 """
52 (nonzero_indices, zero_indices) = _select_indices(instance)
53 xnonzero = numpy.asarray([instance[idx] for idx in nonzero_indices])
17 print 54 print
18 print "xnonzero:", xnonzero 55 print "xnonzero:", xnonzero
19 56
20 (ynonzero, yzero, loss, gw1nonzero, gb1, gw2nonzero, gw2zero, gb2nonzero, gb2zero) = trainfn(xnonzero, self.parameters.w1[nonzero_indexes, :], self.parameters.b1, self.parameters.w2[:, nonzero_indexes], self.parameters.w2[:, zero_indexes], self.parameters.b2[nonzero_indexes], self.parameters.b2[zero_indexes]) 57 (ynonzero, yzero, loss, gw1nonzero, gb1, gw2nonzero, gw2zero, gb2nonzero, gb2zero) = trainfn(xnonzero, self.parameters.w1[nonzero_indices, :], self.parameters.b1, self.parameters.w2[:, nonzero_indices], self.parameters.w2[:, zero_indices], self.parameters.b2[nonzero_indices], self.parameters.b2[zero_indices])
21 print "OLD ynonzero:", ynonzero 58 print "OLD ynonzero:", ynonzero
22 print "OLD yzero:", yzero 59 print "OLD yzero:", yzero
23 print "OLD total loss:", loss 60 print "OLD total loss:", loss
24 61
25 # SGD update 62 # SGD update
26 self.parameters.w1[nonzero_indexes, :] -= LR * gw1nonzero 63 self.parameters.w1[nonzero_indices, :] -= LR * gw1nonzero
27 self.parameters.b1 -= LR * gb1 64 self.parameters.b1 -= LR * gb1
28 self.parameters.w2[:, nonzero_indexes] -= LR * gw2nonzero 65 self.parameters.w2[:, nonzero_indices] -= LR * gw2nonzero
29 self.parameters.w2[:, zero_indexes] -= LR * gw2zero 66 self.parameters.w2[:, zero_indices] -= LR * gw2zero
30 self.parameters.b2[nonzero_indexes] -= LR * gb2nonzero 67 self.parameters.b2[nonzero_indices] -= LR * gb2nonzero
31 self.parameters.b2[zero_indexes] -= LR * gb2zero 68 self.parameters.b2[zero_indices] -= LR * gb2zero
32 69
33 # Recompute the loss, to make sure it's descreasing 70 # Recompute the loss, to make sure it's descreasing
34 (ynonzero, yzero, loss, gw1nonzero, gb1, gw2nonzero, gw2zero, gb2nonzero, gb2zero) = trainfn(xnonzero, self.parameters.w1[nonzero_indexes, :], self.parameters.b1, self.parameters.w2[:, nonzero_indexes], self.parameters.w2[:, zero_indexes], self.parameters.b2[nonzero_indexes], self.parameters.b2[zero_indexes]) 71 (ynonzero, yzero, loss, gw1nonzero, gb1, gw2nonzero, gw2zero, gb2nonzero, gb2zero) = trainfn(xnonzero, self.parameters.w1[nonzero_indices, :], self.parameters.b1, self.parameters.w2[:, nonzero_indices], self.parameters.w2[:, zero_indices], self.parameters.b2[nonzero_indices], self.parameters.b2[zero_indices])
35 print "NEW ynonzero:", ynonzero 72 print "NEW ynonzero:", ynonzero
36 print "NEW yzero:", yzero 73 print "NEW yzero:", yzero
37 print "NEW total loss:", loss 74 print "NEW total loss:", loss