Mercurial > pylearn

comparison algorithms/logistic_regression.py @ 503:c7ce66b4e8f4

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Extensions to algorithms, and some cleanup (by defining linear_output result).
author Joseph Turian <turian@gmail.com>
date Wed, 29 Oct 2008 03:29:18 -0400
parents 17945defd813
children b267a8000f92
comparison
equal deleted inserted replaced
502:17945defd813 503:c7ce66b4e8f4
16 16
17 self.w = N.zeros((n_in, n_out)) 17 self.w = N.zeros((n_in, n_out))
18 self.b = N.zeros(n_out) 18 self.b = N.zeros(n_out)
19 self.lr = 0.01 19 self.lr = 0.01
20 self.__hide__ = ['params'] 20 self.__hide__ = ['params']
21 self.input_dimension = n_in
22 self.output_dimension = n_out
21 23
22 class Module_Nclass(module.FancyModule): 24 class Module_Nclass(module.FancyModule):
23 InstanceType = LogRegInstanceType 25 InstanceType = LogRegInstanceType
24 26
25 def __init__(self, x=None, targ=None, w=None, b=None, lr=None, regularize=False): 27 def __init__(self, x=None, targ=None, w=None, b=None, lr=None, regularize=False):
32 self.b = b if b is not None else module.Member(T.dvector()) 34 self.b = b if b is not None else module.Member(T.dvector())
33 self.lr = lr if lr is not None else module.Member(T.dscalar()) 35 self.lr = lr if lr is not None else module.Member(T.dscalar())
34 36
35 self.params = [p for p in [self.w, self.b] if p.owner is None] 37 self.params = [p for p in [self.w, self.b] if p.owner is None]
36 38
37 xent, output = nnet.crossentropy_softmax_1hot( 39 linear_output = T.dot(self.x, self.w) + self.b
38 T.dot(self.x, self.w) + self.b, self.targ) 40
41 (xent, softmax, max_pr, argmax) = nnet.crossentropy_softmax_max_and_argmax_1hot(
42 linear_output, self.targ)
39 sum_xent = T.sum(xent) 43 sum_xent = T.sum(xent)
40 44
41 self.output = output 45 self.softmax = softmax
46 self.argmax = argmax
47 self.max_pr = max_pr
42 self.sum_xent = sum_xent 48 self.sum_xent = sum_xent
49
50 # Softmax being computed directly.
51 softmax_unsupervised = nnet.softmax(linear_output)
52 self.softmax_unsupervised = softmax_unsupervised
43 53
44 #compatibility with current implementation of stacker/daa or something 54 #compatibility with current implementation of stacker/daa or something
45 #TODO: remove this, make a wrapper 55 #TODO: remove this, make a wrapper
46 self.cost = sum_xent 56 self.cost = self.sum_xent
47 self.input = self.x 57 self.input = self.x
58 # TODO: I want to make output = linear_output.
59 self.output = self.softmax_unsupervised
48 60
49 #define the apply method 61 #define the apply method
50 self.pred = T.argmax(T.dot(self.input, self.w) + self.b, axis=1) 62 self.pred = T.argmax(linear_output, axis=1)
51 self.apply = module.Method([self.input], self.pred) 63 self.apply = module.Method([self.input], self.pred)
64
65 self.validate = module.Method([self.input, self.targ], [self.cost, self.argmax, self.max_pr])
66 self.softmax_output = module.Method([self.input], self.softmax_unsupervised)
52 67
53 if self.params: 68 if self.params:
54 gparams = T.grad(sum_xent, self.params) 69 gparams = T.grad(sum_xent, self.params)
55 70
56 self.update = module.Method([self.input, self.targ], sum_xent, 71 self.update = module.Method([self.input, self.targ], sum_xent,