Mercurial > ift6266

comparison deep/convolutional_dae/run_exp.py @ 291:7d1fa2d7721c

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Split out the run_exp method.
author Arnaud Bergeron <abergeron@gmail.com>
date Fri, 26 Mar 2010 18:35:23 -0400
parents
children 8108d271c30c
comparison
equal deleted inserted replaced
290:518589bfee55 291:7d1fa2d7721c
1 from ift6266.deep.convolutional_dae.scdae import *
2
3 class dumb(object):
4 def save(self):
5 pass
6
7 def go(state, channel):
8 from ift6266 import datasets
9 from ift6266.deep.convolutional_dae.sgd_opt import sgd_opt
10 import pylearn, theano, ift6266
11 import pylearn.version
12
13 # params: bsize, pretrain_lr, train_lr, nfilts1, nfilts2, nftils3, nfilts4
14 # pretrain_rounds, noise, mlp_sz
15
16 pylearn.version.record_versions(state, [theano, ift6266, pylearn])
17 # TODO: maybe record pynnet version?
18 channel.save()
19
20 dset = datasets.nist_all(1000)
21
22 nfilts = []
23 if state.nfilts1 != 0:
24 nfilts.append(state.nfilts1)
25 if state.nfilts2 != 0:
26 nfilts.append(state.nfilts2)
27 if state.nfilts3 != 0:
28 nfilts.append(state.nfilts3)
29 if state.nfilts4 != 0:
30 nfilts.append(state.nfilts4)
31
32 fsizes = [(5,5)]*len(nfilts)
33 subs = [(2,2)]*len(nfilts)
34 noise = [state.noise]*len(nfilts)
35
36 pretrain_funcs, trainf, evalf, net = build_funcs(
37 img_size=(32, 32),
38 batch_size=state.bsize,
39 filter_sizes=fsizes,
40 num_filters=nfilts,
41 subs=subs,
42 noise=noise,
43 mlp_sizes=[state.mlp_sz],
44 out_size=62,
45 dtype=numpy.float32,
46 pretrain_lr=state.pretrain_lr,
47 train_lr=state.train_lr)
48
49 pretrain_fs, train, valid, test = massage_funcs(
50 state.bsize, dset, pretrain_funcs, trainf, evalf)
51
52 series = create_series()
53
54 do_pretrain(pretrain_fs, state.pretrain_rounds, series['recons_error'])
55
56 sgd_opt(train, valid, test, training_epochs=100000, patience=10000,
57 patience_increase=2., improvement_threshold=0.995,
58 validation_frequency=2500, series=series, net=net)
59
60 if __name__ == '__main__':
61 st = dumb()
62 st.bsize = 100
63 st.pretrain_lr = 0.01
64 st.train_lr = 0.1
65 st.nfilts1 = 4
66 st.nfilts2 = 4
67 st.nfilts3 = 0
68 st.pretrain_rounds = 500
69 st.noise=0.2
70 st.mlp_sz = 500
71 go(st, dumb())