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
219
|
#!/usr/bin/python3
# -*- coding: utf-8 -*-
# Copyright © 2010, 2012 marmuta <marmvta@gmail.com>
#
# This file is part of Onboard.
#
# Onboard is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# Onboard is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
import sys, random, string, math
import copy
from optparse import OptionParser
import matplotlib.pyplot as plt
import pypredict
from pypredict import *
class Annealing:
def __call__(self, iterations, attenuation, progress):
current = self.init()
current_fitness = self.fitness(current)
best, best_fitness = current, current_fitness
temperature = 1.0
for iteration in range(iterations):
next = self.modify(current, temperature)
next_fitness = self.fitness(next)
progress(next, next_fitness, best, best_fitness)
if next_fitness > current_fitness:
best, best_fitness = next, next_fitness
current, current_fitness = next, next_fitness
print "accepted: ", current_fitness, current
temperature *= attenuation
return (best_fitness, best)
class Scenario:
def __repr__(self):
return "recency_smoothing=%d recency_ratio=%.3f " \
"recency_halflife=%.0f recency_lambdas=[%s]" % \
(self.recency_smoothing, self.recency_ratio,
self.recency_halflife,
", ".join("%.3f" % l for l in self.recency_lambdas))
def random_step(value, minval, maxval, temperature):
d = (maxval - minval) * temperature
a = max(minval, value - d)
b = min(maxval, value + d)
return random.random() * (b - a) + a
class KsrAnnealing(Annealing):
def init(self):
self.cache = {}
s = Scenario()
s.recency_smoothing = 0
s.recency_ratio = 0.58
s.recency_halflife = 80
s.recency_lambdas = [0.3, 0.3, 0.3]
s.recency_ratio = 0.813
s.recency_halflife = 97
s.recency_lambdas = [0.155, 0.147, 0.411]
#ksr= 26.0088272383 recency_smoothing=0 recency_ratio=0.784 recency_halflife=78 recency_lambdas=[0.154, 0.658, 0.270]
s.recency_ratio = 0.784
s.recency_halflife = 78
s.recency_lambdas = [0.154, 0.658, 0.270]
#ksr= 26.013126218 recency_smoothing=0 recency_ratio=0.811 recency_halflife=96 recency_lambdas=[0.404, 0.831, 0.444]
s.recency_ratio=0.811
s.recency_halflife=96
s.recency_lambdas=[0.404, 0.831, 0.444]
#ksr= 26.0145592113 recency_smoothing=0 recency_ratio=0.811 recency_halflife=96 recency_lambdas=[0.159, 0.967, 0.036]
s.recency_ratio=0.811
s.recency_halflife=96
s.recency_lambdas=[0.159, 0.967, 0.036]
s.recency_smoothing = 1
if 0:
s.recency_smoothing = 0
s.recency_ratio = random.random()
s.recency_halflife = random.randrange(1,200)
s.recency_lambdas = [random.random(), random.random(), random.random()]
return s
def modify(self, scenario, temperature):
temperature *= 2
s = copy.copy(scenario)
s.recency_ratio = random_step(s.recency_ratio, 0, 1.0, temperature)
s.recency_halflife = random_step(s.recency_halflife, 1, 200, temperature)
for i in range(3):
s.recency_lambdas[i] = random_step(s.recency_lambdas[i], 0, 1.0, temperature)
return s
def fitness(self, scenario):
s = scenario
v = (s.recency_smoothing, s.recency_ratio, s.recency_halflife,
tuple(s.recency_lambdas))
if v in self.cache:
return self.cache[v]
for m in models:
m.smoothing = "abs-disc"
model = overlay(models)
learn_model = models[1]
learn_model.clear()
learn_model.recency_ratio = s.recency_ratio
learn_model.recency_halflife = s.recency_halflife
index = min(max(int(s.recency_smoothing),0),len(recency_smoothings)-1)
learn_model.recency_smoothing = recency_smoothings[index]
learn_model.lambdas = s.recency_lambdas
ksr = pypredict.ksr(model, learn_model, testing_sentences, 10)#,
#lambda i,n, c, p: sys.stdout.write("%d/%d\n" % (i+1,n)))
#ksr = -((s.recency_ratio-.2)**2) + 26
self.cache[v] = ksr
return ksr
class PlotProgress:
def __init__(self):
self.xvalues = []
self.ksrs = []
self.best_ksrs = []
def __call__(self, scenario, ksr, best_scenario, best_ksr):
print "best: ksr=", best_ksr, best_scenario
print "current: ksr=", ksr, scenario
self.xvalues.append(len(self.xvalues))
self.ksrs.append(ksr)
self.best_ksrs.append(best_ksr)
plt.ion() # interactive mode on
plt.figure(1)
plt.clf()
lines = [
plt.plot(self.xvalues, self.ksrs),
plt.plot(self.xvalues, self.best_ksrs),
]
labels = ["current", "best"]
plt.xlabel("iteration")
plt.ylabel('ksr [%]')
ymin, ymax = plt.ylim()
plt.ylim(ymin, ymax+(ymax-ymin)*0.05)
plt.figlegend(lines, labels, 'upper right')
plt.gcf().suptitle('Optimizing Recency Caching',
fontsize=16)
plt.draw()
def optimize_caching(base_model, testing, order):
global models, testing_text, testing_sentences, testing_tokens
global recency_smoothings
models = []
recency_smoothings = ["jelinek-mercer", "witten-bell"]
with timeit("loading base model '%s'" % (base_model,)):
model = DynamicModelKN(order)
model.load(base_model)
models.append(model)
model = CachedDynamicModel(order)
models.append(model)
filename = testing
with timeit("tokenizing '%s'" % (filename,)):
testing_text = read_corpus(filename)
testing_sentences = split_sentences(testing_text)
testing_tokens, spans = tokenize_text(testing_text)
# simulated annealing
print KsrAnnealing()(200, 0.95, PlotProgress())
plt.show() # blocks; allows for interaction with the chart, saving images
def main():
parser = OptionParser(usage= \
"""Usage: %prog [options] caching <base model> <testing text>
<base model> is a static base language model
<testing text> is simulated input that incrementally
trains the second language model""")
options, args = parser.parse_args()
if len(args) < 1:
parser.print_usage()
sys.exit(1)
order = 3
command = args[0].lower()
if command == "caching":
if len(args) < 3:
parser.print_usage()
sys.exit(1)
optimize_caching(args[1], args[2], order)
else:
print "unknown command '%s', exiting" % command
sys.exit(1)
if __name__ == '__main__':
main()
|
