/
mnistExample.py
76 lines (62 loc) · 2.88 KB
/
mnistExample.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
"""
Copyright (c) 2011,2012 George Dahl
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject
to the following conditions:
The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software. THE
SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
"""
import numpy as num
import itertools
from gdbn.dbn import *
def numMistakes(targetsMB, outputs):
if not isinstance(outputs, num.ndarray):
outputs = outputs.as_numpy_array()
if not isinstance(targetsMB, num.ndarray):
targetsMB = targetsMB.as_numpy_array()
return num.sum(outputs.argmax(1) != targetsMB.argmax(1))
def sampleMinibatch(mbsz, inps, targs):
idx = num.random.randint(inps.shape[0], size=(mbsz,))
return inps[idx], targs[idx]
def main():
mbsz = 64
layerSizes = [784, 512, 512, 10]
scales = [0.05 for i in range(len(layerSizes)-1)]
fanOuts = [None for i in range(len(layerSizes)-1)]
learnRate = 0.1
epochs = 10
mbPerEpoch = int(num.ceil(60000./mbsz))
f = num.load("mnist.npz")
trainInps = f['trainInps']/255.
testInps = f['testInps']/255.
trainTargs = f['trainTargs']
testTargs = f['testTargs']
assert(trainInps.shape == (60000, 784))
assert(trainTargs.shape == (60000, 10))
assert(testInps.shape == (10000, 784))
assert(testTargs.shape == (10000, 10))
mbStream = (sampleMinibatch(mbsz, trainInps, trainTargs) for unused in itertools.repeat(None))
net = buildDBN(layerSizes, scales, fanOuts, Softmax(), False)
net.learnRates = [learnRate for x in net.learnRates]
net.L2Costs = [0 for x in net.L2Costs]
net.nestCompare = True #this flag existing is a design flaw that I might address later, for now always set it to True
for ep, (trCE, trEr) in enumerate(net.fineTune(mbStream, epochs, mbPerEpoch, numMistakes, True)):
print ep, trCE, trEr
outputs = tuple(net.predictions(testInps))
outputs = num.array(outputs).reshape(testInps.shape[0], -1)
print "Test error rate:", numMistakes(
testTargs, outputs) / float(testInps.shape[0])
if __name__ == "__main__":
main()