-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmetrics.py
218 lines (178 loc) · 6.54 KB
/
metrics.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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
import numpy as np
import util
import scipy.stats
import sdm
import hopfield as hop
from joblib import Memory
memory = Memory(cachedir="cache", mmap_mode='c', verbose=0)
@memory.cache
def test_hopfield_capacity_n(n, k=1, iters=100):
corruption = np.empty((iters, k))
# store the same number of items multiple times
for i in xrange(iters):
# generate random inputs
vecs = util.random_input(n, k)
# create hopfield net
mem = hop.hopnet(vecs)
# read the items backout
r = mem.readM(vecs, 1000)
# find the largest fraction of corrupted bits
corruption[i] = np.mean(r ^ vecs, axis=0)
return corruption
@memory.cache
def test_sdm_capacity_n(params, k=1, iters=100):
n, m, D = params
mem = sdm.SDM(n, m, D)
corruption = np.empty((iters, k))
# store the same number of items multiple times
for i in xrange(iters):
# generate random inputs
vecs = util.random_input(n, k)
# reset the memory to its original state
mem.reset()
# write random inputs to memory
mem.writeM(vecs, vecs)
# read the items back out
r = mem.readM(vecs)
# find the largest fraction of corrupted bits
corruption[i] = np.mean(r ^ vecs, axis=0)
return corruption
def test_capacity(params, k0, iters=100, verbose=False):
if hasattr(params, '__iter__'):
testfunc = test_sdm_capacity_n
else:
testfunc = test_hopfield_capacity_n
data = np.empty((len(k0), 2))
# test storing different numbers of items
for kidx, k in enumerate(k0):
# compute capacity
corruption = testfunc(
params, k=int(k), iters=iters)
# compute statistics about the distances
maxc = np.mean(corruption, axis=1)
mean = np.mean(maxc)
sem = scipy.stats.sem(maxc)
data[kidx] = (mean, sem)
if verbose:
print "%2d: %.3f +/- %.3f" % (k, mean, sem)
return data
######################################################################
@memory.cache
def test_hopfield_noise_tolerance_n(n, k=1, noise=0, iters=100):
if noise == 0:
return test_hopfield_capacity_n(n, k=k, iters=iters)
corruption = np.empty((iters, k))
bits = int(n * noise)
# store the same number of items multiple times
for i in xrange(iters):
# generate random inputs
vecs = util.random_input(n, k)
cvecs = util.corrupt(vecs, bits)
# create hopfield net
mem = hop.hopnet(vecs)
# read the items backout
r = mem.readM(cvecs, 1000)
# find the largest fraction of corrupted bits
corruption[i] = np.mean(r ^ vecs, axis=0)
return corruption
@memory.cache
def test_sdm_noise_tolerance_n(params, k=1, noise=0, iters=100):
if noise == 0:
return test_sdm_capacity_n(params, k=k, iters=iters)
n, m, D = params
mem = sdm.SDM(n, m, D)
corruption = np.empty((iters, k))
bits = int(n * noise)
# store the same number of items multiple times
for i in xrange(iters):
# generate random inputs
vecs = util.random_input(n, k)
cvecs = util.corrupt(vecs, bits)
# reset the memory to its original state
mem.reset()
# write random inputs to memory
mem.writeM(vecs, vecs)
# read the items back out
r = mem.readM(cvecs)
# find the largest fraction of corrupted bits
corruption[i] = np.mean(r ^ vecs, axis=0)
return corruption
def test_noise_tolerance(params, k0, noise=0, iters=100, verbose=False):
if hasattr(params, '__iter__'):
testfunc = test_sdm_noise_tolerance_n
else:
testfunc = test_hopfield_noise_tolerance_n
data = np.empty((len(k0), 2))
# test storing different numbers of items
for kidx, k in enumerate(k0):
# compute noise tolerance
corruption = testfunc(
params, k=int(k), noise=noise, iters=iters)
# compute statistics about the distances
maxc = np.mean(corruption, axis=1)
mean = np.mean(maxc)
sem = scipy.stats.sem(maxc)
data[kidx] = (mean, sem)
if verbose:
print "%2d: %.3f +/- %.3f" % (k, mean, sem)
return data
######################################################################
@memory.cache
def test_hopfield_prototype_n(n, kp=1, ke=1, noise=0, iters=100):
corruption = np.empty((iters, kp))
bits = int(n * noise)
# store the same number of items multiple times
for i in xrange(iters):
# generate random inputs
vecs = util.random_input(n, kp)
cvecs = vecs[..., None] * np.ones((n, kp, ke), dtype='i4')
cvecs = util.corrupt(
cvecs.reshape((n, kp*ke)),
bits, with_replacement=True)
ex = util.corrupt(vecs, bits)
# create hopfield net
mem = hop.hopnet(cvecs)
# read the items backout
r = mem.readM(ex, 1000)
# find the largest fraction of corrupted bits
corruption[i] = np.mean(r ^ vecs, axis=0)
return corruption
@memory.cache
def test_sdm_prototype_n(params, kp=1, ke=1, noise=0, iters=100):
n, m, D = params
mem = sdm.SDM(n, m, D)
corruption = np.empty((iters, kp))
bits = int(n * noise)
# store the same number of items multiple times
for i in xrange(iters):
# generate random prototype and exemplars
vecs = util.random_input(n, kp)
cvecs = vecs[..., None] * np.ones((n, kp, ke), dtype='i4')
cvecs = util.corrupt(
cvecs.reshape((n, kp*ke)),
bits, with_replacement=True)
ex = util.corrupt(vecs, bits)
# reset the memory to its original state
mem.reset()
# write random inputs to memory
mem.writeM(cvecs, cvecs)
# read the items back out
r = mem.readM(ex)
# find the fraction of corrupted bits
corruption[i] = np.mean(r ^ vecs, axis=0)
return corruption
def test_prototype(params, kp=1, ke=1, noise=0, iters=100, verbose=False):
if hasattr(params, '__iter__'):
testfunc = test_sdm_prototype_n
else:
testfunc = test_hopfield_prototype_n
# compute noise tolerance
corruption = testfunc(
params, kp=kp, ke=ke, noise=noise, iters=iters)
# compute statistics about the distances
maxc = np.mean(corruption, axis=1)
mean = np.mean(maxc)
sem = scipy.stats.sem(maxc)
if verbose:
print "%2d %2d: %.3f +/- %.3f" % (kp, ke, mean, sem)
return mean