/
fwk4xps.py
521 lines (411 loc) · 16.2 KB
/
fwk4xps.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
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
import ConfigParser
import subprocess
import multiprocessing
import itertools
from thread import start_new_thread, allocate_lock
from multiprocessing import Process, Value, Lock
import ctypes
import numpy as np
import copy_reg
import types
import random
import time
import numpy as np
import math
from random import shuffle
from scipy.stats import t
from scipy import stats
import os
lock = Lock()
shared_times = []
shared_boxes = []
shared_gains = []
shared_run = []
shared_avgtime = []
shared_stddev = []
shared_runinst = []
shared_L = []
shared_W = []
shared_stop = 0
class Algo:
def __init__(self, configParser, name):
self.name = name
if configParser.has_option(name, 'output_dir'):
self.output_dir = configParser.get(name, 'output_dir')
else:
self.output_dir = name
self.ex = configParser.get(name, 'exec')
self.params = configParser.get(name, 'params')
self.params = self.params.replace('_ALGO', self.ex);
self.params = self.params.replace('_MAXTIME', configParser.get('base', 'maxtime'));
self.outputpos_times = configParser.get(name, 'outputpos_times')
self.outputpos_boxes = configParser.get(name, 'outputpos_boxes')
def run(self, inst, seed):
run_exec = self.ex + " " + self.params.replace('_INSTANCE', inst)
run_exec = run_exec.replace('_SEED', str(seed))
print run_exec
output = subprocess.check_output(run_exec,shell=True,).splitlines()[-1]
try:
return (float(output.split()[int(self.outputpos_times)]), float(output.split()[int(self.outputpos_boxes)]))
except ValueError:
print "error with instance " + inst.split('/')[-1] + "\n" + output
return -2.0
class Config:
def __init__(self, filename):
configParser = ConfigParser.RawConfigParser()
configParser.read(filename)
self.instances = configParser.get('base', 'test_instances')
self.mintime = float(configParser.get('base', 'mintime'))
self.maxtime = float(configParser.get('base', 'maxtime'))
self.sig_gain = float(configParser.get('base', 'sig_gain'))
self.max_seeds = int(configParser.get('base', 'max_seeds'))
self.min_seeds = int(configParser.get('base', 'min_seeds'))
f = open(self.instances)
self.instances = f.read().splitlines()
f.close()
self.idx2inst = range(len(self.instances))
for i in range(len(self.instances)): self.idx2inst[i]=i
random.seed(0)
shuffle(self.idx2inst)
self.algos = []
for algo_name in configParser.get('base', 'algo_names').split():
if configParser.has_option(algo_name, 'tuning_param'):
tuning_param=configParser.get(algo_name, 'tuning_param')
for value in configParser.get(algo_name, 'tuning_values').split():
algo=Algo(configParser,algo_name)
self.algos.append(algo)
algo.params = algo.params.replace('_%s'%tuning_param, value);
algo.name = algo.name + "_" + tuning_param + value
else:
algo=Algo(configParser,algo_name)
self.algos.append(algo)
def onKeyPress(event):
global shared_stop
shared_stop=1
def read_times(name,i,config):
global shared_times, shared_boxes
shared_times[i,:] = -1.0
shared_boxes[i,:] = -1.0
try:
f = open("%(dir)s/%(name)s_times.out" % {"dir":config.algos[i].output_dir,"name":name}, )
f2 = open("%(dir)s/%(name)s_boxes.out" % {"dir":config.algos[i].output_dir,"name":name}, )
idx = 0
for lines in f:
#~ print lines.split()
seed=0
for line in lines.split():
if seed==config.max_seeds: break
shared_times[i,idx + len(config.instances)*seed ] = float(line)
if shared_times[i,idx + len(config.instances)*seed]==-1.5: shared_times[i,idx + len(config.instances)*seed]=-1.0
seed += 1
idx+=1
idx = 0
for lines in f2:
#~ print lines.split()
seed=0
for line in lines.split():
if seed==config.max_seeds: break
shared_boxes[i,idx + len(config.instances)*seed ] = float(line)
if shared_boxes[i,idx + len(config.instances)*seed]==-1.5: shared_boxes[i,idx + len(config.instances)*seed]=-1.0
seed += 1
idx+=1
#~ print shared_times[i]
f.close()
f2.close()
except IOError:
None
timesA = []
for elem in (j for j in shared_times[i,:] if j > 0.0):
timesA.append(elem)
#shared_run[i]=len(timesA)
def write_times(name,i,config):
global shared_times, shared_boxes
try:
os.mkdir(config.algos[i].output_dir)
except OSError:
None
f = open("%(dir)s/%(name)s_times.out" % {"dir":config.algos[i].output_dir,"name":name}, "w")
f2 = open("%(dir)s/%(name)s_meantimes.out" % {"dir":config.algos[i].output_dir,"name":name}, "w")
for j in range(len(config.instances)):
for k in range(config.max_seeds):
f.write("%.2f " % shared_times[i][k*len(config.instances)+j])
f.write("\n")
(mean,h,n) = mean_error(config, i, j, config.max_seeds)
f2.write("%.2f " % mean)
f2.write("+- %.2f" % h)
f2.write(" (%d)\n" % n)
f.close()
f2.close()
f = open("%(dir)s/%(name)s_boxes.out" % {"dir":config.algos[i].output_dir,"name":name}, "w")
f2 = open("%(dir)s/%(name)s_meanboxes.out" % {"dir":config.algos[i].output_dir,"name":name}, "w")
for j in range(len(config.instances)):
for k in range(config.max_seeds):
f.write("%.2f " % shared_boxes[i][k*len(config.instances)+j])
f.write("\n")
(mean,h,n) = mean_error(config, i, j, config.max_seeds)
f2.write("%.2f " % mean)
f2.write("+- %.2f" % h)
f2.write(" (%d)\n" % n)
f.close()
f2.close()
def update_algo(config,id_algo):
shared_nb_comp[id_algo] =0
shared_L[id_algo] = 0
shared_W[id_algo] = 0
fL = open("%(dir)s/%(name)s_badresults.out" % {"dir":config.algos[id_algo].output_dir,"name":config.algos[id_algo].name}, "w")
fW = open("%(dir)s/%(name)s_goodresults.out" % {"dir":config.algos[id_algo].output_dir,"name":config.algos[id_algo].name}, "w")
for i in (k for k in range(len(config.instances)) if shared_times[id_algo,k] >= 0.0 and shared_times[0,k]>=0.0):
if config.instances[i].endswith('*'): continue
n=0.;sum0=0.;sum1=0.
for kk in range(config.max_seeds):
if shared_times[0,i+kk*len(config.instances)] >= 0.0 and shared_times[id_algo,i+kk*len(config.instances)]>=0.0:
sum0+=shared_times[0,i+kk*len(config.instances)]
sum1+=shared_times[id_algo,i+kk*len(config.instances)]
n+=1.
if n>0 and (sum0/n>config.mintime or sum1/n>config.mintime) and (sum0/n<config.maxtime or sum1/n<config.maxtime):
if sum0/sum1 > float(config.sig_gain):
shared_W[id_algo]+=1
fW.write("%s %f %f\n" % (config.instances[i],sum1/n,sum0/n))
if sum1/sum0 > float(config.sig_gain):
shared_L[id_algo]+=1
fL.write("%s %f %f\n" % (config.instances[i],sum1/n,sum0/n))
shared_nb_comp[id_algo] +=1
if shared_nb_comp[id_algo] > 0:
av=av_rel_time(id_algo, config)
shared_gains[id_algo]=(1.-av)/av
shared_gains_ttime[id_algo]=gain_total_time(id_algo, config)
else:
shared_gains[id_algo]=1.0
shared_gains_ttime[id_algo]=1.0
fW.close()
fL.close()
def update_all(config):
for id_algo in range(len(config.algos)):
update_algo(config,id_algo)
#~ for id_inst in range(len(config.instances)):
#~ update_inst(config,id_inst)
print shared_gains
print shared_run
print shared_nb_comp
def update_share(config,id_algo,id_inst):
#~ update_inst(config,id_inst)
update_algo(config,id_algo)
f = open("state.out", "w")
f.write("algos:")
for elem in config.algos:
f.write("%s " % elem.name)
f.write("\n")
f.write("gains (average relative times):")
for elem in shared_gains:
f.write("%f " % elem)
f.write("\n")
f.write("gains (total time) :")
for elem in shared_gains_ttime:
f.write("%f " % elem)
f.write("\n")
f.write("compared instances:")
for elem in shared_nb_comp:
f.write("%3d " % elem)
f.write("\n")
f.write("runs :")
for elem in shared_run:
f.write("%3d " % elem)
f.write("\n")
f.write("L:")
for elem in shared_L:
f.write("%3d " % elem)
f.write("\n")
f.write("W:")
for elem in shared_W:
f.write("%3d " % elem)
f.write("\n")
f.close()
def av_rel_time(i,config):
if i==0: return 0.5
x=[]
for k in range(len(config.instances)):
if config.instances[k].endswith('*'): continue
n0=0;n1=0;sum0=0;sum1=0; sum2=0
for kk in range(config.max_seeds):
if shared_times[0,k+kk*len(config.instances)] >=0.0:
sum0+=shared_times[0,k+kk*len(config.instances)]
n0+=1
if shared_times[i,k+kk*len(config.instances)] >=0.0:
sum1+=shared_times[i,k+kk*len(config.instances)]
n1+=1
if n0>0 and n1>0 and (sum0/float(n0)>config.mintime or sum1/float(n1)>config.mintime) and (sum0/float(n0)<config.maxtime or sum1/float(n1)<config.maxtime):
rel1 = (sum1/float(n1)) /(sum1/float(n1)+sum0/float(n0)+0.001)
x.append(rel1)
xmean=np.mean(x)
return xmean
def gain_total_time(i,config):
if i==0: return 1.0
total_time0=0.
total_time1=0.
for k in range(len(config.instances)):
if config.instances[k].endswith('*'): continue
n0=0.; n1=0. ;sum0=0.;sum1=0.
for kk in range(config.max_seeds):
if shared_times[0,k+kk*len(config.instances)] >=0.0:
sum0+=shared_times[0,k+kk*len(config.instances)]
n0+=1.
if shared_times[i,k+kk*len(config.instances)] >=0.0:
sum1+=shared_times[i,k+kk*len(config.instances)]
n1+=1.
if n0>0. and n1>0. and ((sum0/n0)<config.maxtime or (sum1/n1)<config.maxtime):
total_time0+=(sum0/n0)
total_time1+=(sum1/n1)
return total_time0/(total_time1+0.01)
def p_faster(i, j,config):
if i==j: return 0.5
x=[]
for k in range(len(config.instances)):
if config.instances[k].endswith('*'): continue
n=0; sum0=0; sum1=0; sum2=0
for kk in range(config.max_seeds):
if shared_times[0,k+kk*len(config.instances)] >=0.0 and shared_times[i,k+kk*len(config.instances)] >=0.0 and shared_times[j,k+kk*len(config.instances)] >=0.0:
sum0+=shared_times[0,k+kk*len(config.instances)]
sum1+=shared_times[i,k+kk*len(config.instances)]
sum2+=shared_times[j,k+kk*len(config.instances)]
n+=1
if n>0 and (sum0/float(n)>config.mintime or sum1/float(n)>config.mintime or sum2/float(n)>config.mintime) and (sum0/float(n)<config.maxtime or sum1/float(n)<config.maxtime or sum2/float(n)<config.maxtime):
rel1=(sum1)/(sum1+sum0+0.001)
rel2=(sum2)/(sum2+sum0+0.001)
x.append(rel2-rel1)
#x.append(1.0) #the best x for algo i (only for seed 1?), then worst_time/seed
n= len(x)
if n<=1: return 1.0
return t.cdf((np.mean(x))/(np.std(x)/math.sqrt(n)),n-1)
def mean_error(config, id_algo, id_inst, seeds):
x=[]
for kk in range(seeds):
if shared_times[id_algo,id_inst+kk*len(config.instances)] >=0.0:
x.append(shared_times[id_algo,id_inst+kk*len(config.instances)])
n=len(x)
if n>=2:
se=stats.sem(x)
h = se*t._ppf((1+0.95)/2., n-1)
return (np.mean(x),h,n)
else:
return (np.mean(x),-1,n)
def next_run(config):
global lock
global shared_stop
lock.acquire()
if shared_stop==1: return None
for i in range(len(config.algos)):
idx = shared_run[i]%len(config.instances)
seed = int(shared_run[i]/len(config.instances)) + 1
while shared_run[i] < len(config.instances)*config.max_seeds:
if config.instances[config.idx2inst[idx]].endswith('*'):
shared_run[i]+=1
idx = shared_run[i]%len(config.instances)
seed = int(shared_run[i]/len(config.instances)) + 1
continue
#en caso de que el algo_instancia se hayan corrido mas de min_seed veces, se calcula el diametro de confianza para el tiempo medio
#si el diametro es < 0.1 no se sigue corriendo el par
if seed>config.min_seeds and shared_times[i,len(config.instances)*(seed-1)+config.idx2inst[idx]] == -1.0:
(mean,h,n)=mean_error(config,i,config.idx2inst[idx],seed-1)
if mean>0:
h/=mean
if n>=config.min_seeds and h<0.1: shared_times[i,len(config.instances)*(seed-1)+config.idx2inst[idx]]=-1.6
if shared_times[i,len(config.instances)*(seed-1)+config.idx2inst[idx]] != -1.0:
shared_run[i]+=1
idx = shared_run[i]%len(config.instances)
seed = int(shared_run[i]/len(config.instances)) + 1
else: break
if shared_run.min() == len(config.instances)*config.max_seeds:
lock.release()
return None
if shared_run[0]<shared_run.max():
id_algo=0
else:
if shared_run.min() < 20:
id_algo=shared_run.argmin()
else:
#se selecciona el algoritmo con menor tiempo promedio
sorted_idx = sorted(range(len(config.algos)), key=lambda k: av_rel_time(k,config))
print "gains:", shared_gains
for id_algo in (j for j in sorted_idx if shared_run[j] < len(config.instances)*config.max_seeds):
break
#se calcula la probabilidad, para cada algoritmo, de tener un tiempo promedio menor al seleccionado
choices=[]
for i in range(len(config.algos)):
if shared_run[i] < len(config.instances)*config.max_seeds:
choices.append(math.pow(p_faster(i,id_algo,config),0.5))
else :
choices.append(0.0)
#se escoge usando la ruleta segun las probabilidades obtenidas
print choices
max=sum(choices)
pick=random.uniform(0,max)
current=0
id_algo2=0
for value in choices:
current+=value
if current>pick:
break
id_algo2+=1
#el algoritmo con menor promedio es escogido si ha sido corrido un menor o igual numero de veces
print config.algos[id_algo].name," ", config.algos[id_algo2].name
if shared_run[id_algo2] < shared_run[id_algo]: id_algo=id_algo2
idx = shared_run[id_algo]%len(config.instances)
seed = int(shared_run[id_algo]/len(config.instances)) + 1
shared_times[id_algo,len(config.instances)*(seed-1)+config.idx2inst[idx]]=-1.5
shared_run[id_algo] += 1
lock.release()
inst=config.instances[config.idx2inst[idx]]
(output_time, output_box) = config.algos[id_algo].run(inst,seed)
lock.acquire()
if shared_stop==1: return None
shared_times[id_algo,len(config.instances)*(seed-1)+config.idx2inst[idx]] = output_time
shared_boxes[id_algo,len(config.instances)*(seed-1)+config.idx2inst[idx]] = output_box
print config.algos[id_algo].name, inst, output_time
update_share(config,id_algo,len(config.instances)*(seed-1)+config.idx2inst[idx])
write_times(config.algos[id_algo].name,id_algo,config)
lock.release()
if __name__ == '__main__':
config = Config('config.txt')
max_algos = len(config.algos)
shared_times_base = multiprocessing.Array(ctypes.c_float, len(config.instances)*max_algos*config.max_seeds)
shared_times = np.ctypeslib.as_array(shared_times_base.get_obj())
shared_times = shared_times.reshape(max_algos, len(config.instances)* config.max_seeds)
shared_boxes_base = multiprocessing.Array(ctypes.c_float, len(config.instances)*max_algos*config.max_seeds)
shared_boxes = np.ctypeslib.as_array(shared_boxes_base.get_obj())
shared_boxes = shared_boxes.reshape(max_algos, len(config.instances)* config.max_seeds)
shared_gains_base = multiprocessing.Array(ctypes.c_float, max_algos)
shared_gains = np.ctypeslib.as_array(shared_gains_base.get_obj())
shared_gains_ttime_base = multiprocessing.Array(ctypes.c_float, max_algos)
shared_gains_ttime = np.ctypeslib.as_array(shared_gains_ttime_base.get_obj())
shared_run_base = multiprocessing.Array(ctypes.c_int, max_algos)
shared_run = np.ctypeslib.as_array(shared_run_base.get_obj())
shared_L_base = multiprocessing.Array(ctypes.c_int, max_algos)
shared_L = np.ctypeslib.as_array(shared_L_base.get_obj())
shared_W_base = multiprocessing.Array(ctypes.c_int, max_algos)
shared_W = np.ctypeslib.as_array(shared_W_base.get_obj())
shared_runinst_base = multiprocessing.Array(ctypes.c_int, len(config.instances))
shared_runinst = np.ctypeslib.as_array(shared_runinst_base.get_obj())
shared_nb_comp_base = multiprocessing.Array(ctypes.c_int, max_algos)
shared_nb_comp = np.ctypeslib.as_array(shared_nb_comp_base.get_obj())
shared_avgtime_base = multiprocessing.Array(ctypes.c_float, len(config.instances))
shared_avgtime = np.ctypeslib.as_array(shared_avgtime_base.get_obj())
shared_stddev_base = multiprocessing.Array(ctypes.c_float, len(config.instances))
shared_stop = multiprocessing.Value('i', 0)
i=0
for algo in config.algos:
read_times(config.algos[i].name,i,config)
write_times(config.algos[i].name,i,config)
i = i+1
num_cores = multiprocessing.cpu_count()
pool = multiprocessing.Pool(num_cores)
#~ pool.map(run_instance, itertools.izip(config.instances,range(len(config.instances)),itertools.repeat(config.algos[0]),itertools.repeat(0)))
update_all(config)
update_share(config,0,0)
#~ next_run(config)
pool.map(next_run, itertools.repeat(config,len(config.instances)*config.max_seeds*len(config.algos)))
i=0
for algo in config.algos:
write_times(config.algos[i].name,i,config)
i = i+1
update_all(config)
#~ next_run(config)