-
Notifications
You must be signed in to change notification settings - Fork 0
/
mnist.py
executable file
·164 lines (144 loc) · 6.51 KB
/
mnist.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
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
#!/usr/bin/env python
import logging
import numpy as np
import theano
from argparse import ArgumentParser
from theano import tensor
from skimage.transform import rotate
from blocks.bricks.cost import CategoricalCrossEntropy, MisclassificationRate
from blocks.bricks import MLP, Identity, Tanh, Softmax, Rectifier
from blocks.bricks.recurrent import SimpleRecurrent
from blocks.algorithms import GradientDescent, Adam, CompositeRule, StepClipping
from blocks.initialization import IsotropicGaussian, Orthogonal, Constant
from fuel.streams import DataStream
from fuel.datasets import MNIST
from fuel.schemes import SequentialScheme
from fuel.transformers import Mapping, Flatten
from blocks.graph import ComputationGraph
from blocks.monitoring import aggregation
from blocks.extensions import FinishAfter, Timing, Printing
from blocks.extensions.monitoring import (DataStreamMonitoring,
TrainingDataMonitoring)
from blocks.main_loop import MainLoop
from blocks_contrib.extensions import DataStreamMonitoringAndSaving
floatX = theano.config.floatX
mnist = MNIST('train', sources=['features'])
handle = mnist.open()
data = mnist.get_data(handle, slice(0, 50000))[0]
means = data.reshape((50000, 784)).mean(axis=0)
def autocorrentropy2(X, ksize=np.inf):
b, t, d = X.shape
V = np.zeros((b, t, d))
for i in range(b):
for j in range(t):
if ksize in (np.inf, 0., np.nan):
V[i, j, :] = (X[i, :(t-j), :] * X[i, j:, :]).sum(axis=0) / (t-j)
else:
V[i, j, :] = np.exp((-ksize * (X[i, :(t-j), :]-X[i, j:, :])**2)).sum(axis=0) / (t-j)
return V
def allrotations(image, N):
angles = np.linspace(0, 350, N)
R = np.zeros((N, 784))
for i in xrange(N):
img = rotate(image, angles[i])
R[i] = img.flatten()
return R
def _meanize(n_steps, flag=False, ksize=np.inf):
def func(data):
newfirst = data[0] - means[None, :]
Rval = np.zeros((n_steps, newfirst.shape[0], newfirst.shape[1]))
for i, sample in enumerate(newfirst):
Rval[:, i, :] = allrotations(sample.reshape((28, 28)), n_steps)
Rval[10:, i, :] = 0.
num = np.random.randint(0, n_steps, size=1)
Rval[:, i, :] = np.roll(Rval[:, i, :], num, axis=0)
# Rval = newfirst[np.newaxis].repeat(n_steps, axis=0)
if flag:
V = autocorrentropy2(Rval.transpose(1, 0, 2), ksize=ksize)
Rval = V.transpose(1, 0, 2)
Rval = Rval.astype(floatX)
return (Rval, data[1])
return func
def main(save_to, num_epochs, flag, ksize):
batch_size = 128
dim = 100
n_steps = 20
i2h1 = MLP([Identity()], [784, dim], biases_init=Constant(0.), weights_init=IsotropicGaussian(.001))
h2o1 = MLP([Rectifier(), Softmax()], [dim, dim, 10], biases_init=Constant(0.), weights_init=IsotropicGaussian(.001))
rec1 = SimpleRecurrent(dim=dim, activation=Tanh(), weights_init=Orthogonal())
i2h1.initialize()
h2o1.initialize()
rec1.initialize()
x = tensor.tensor3('features')
y = tensor.lmatrix('targets')
preproc = i2h1.apply(x)
h1 = rec1.apply(preproc)
probs = tensor.flatten(h2o1.apply(h1[-1],), outdim=2)
cost = CategoricalCrossEntropy().apply(y.flatten(), probs)
error_rate = MisclassificationRate().apply(y.flatten(), probs)
cost.name = 'final_cost'
error_rate.name = 'error_rate'
cg = ComputationGraph([cost, error_rate])
mnist_train = MNIST("train", subset=slice(0, 50000))
mnist_valid = MNIST("train", subset=slice(50000, 60000))
mnist_test = MNIST("test")
trainstream = Mapping(Flatten(DataStream(mnist_train,
iteration_scheme=SequentialScheme(50000, batch_size))),
_meanize(n_steps, flag, ksize))
validstream = Mapping(Flatten(DataStream(mnist_valid,
iteration_scheme=SequentialScheme(10000,
batch_size))),
_meanize(n_steps, flag, ksize))
teststream = Mapping(Flatten(DataStream(mnist_test,
iteration_scheme=SequentialScheme(10000,
batch_size))),
_meanize(n_steps, flag, ksize))
algorithm = GradientDescent(
cost=cost, params=cg.parameters,
step_rule=CompositeRule([Adam(), StepClipping(100)]))
main_loop = MainLoop(
algorithm,
trainstream,
extensions=[Timing(),
FinishAfter(after_n_epochs=num_epochs),
DataStreamMonitoring(
[cost, error_rate],
validstream,
prefix="test"),
DataStreamMonitoringAndSaving(
[cost, error_rate],
teststream,
[i2h1, h2o1, rec1],
'best_'+save_to+'.pkl',
cost_name=error_rate.name,
after_epoch=True,
prefix='valid'
),
TrainingDataMonitoring(
[cost,
aggregation.mean(algorithm.total_gradient_norm)],
prefix="train",
after_epoch=True),
# Plot(
# save_to,
# channels=[
# ['test_final_cost',
# 'test_misclassificationrate_apply_error_rate'],
# ['train_total_gradient_norm']]),
Printing()])
main_loop.run()
if __name__ == "__main__":
logging.basicConfig(level=logging.INFO)
parser = ArgumentParser("An example of training an MLP on"
" the MNIST dataset.")
parser.add_argument("--num-epochs", type=int, default=2,
help="Number of training epochs to do.")
parser.add_argument("save_to", default="mnist.pkl", nargs="?",
help=("Destination to save the state of the training "
"process."))
parser.add_argument("--flag", type=bool, default=False,
help="Use autocorrentropy or not?")
parser.add_argument("--ksize", type=float, default=np.inf,
help="Kernel size of the autocorrentropy function.")
args = parser.parse_args()
main(args.save_to, args.num_epochs, args.flag, args.ksize)