summaryrefslogtreecommitdiff
path: root/training/mira/mira.py
blob: c84a8cff216c49172e52c811317bc22394022ab4 (plain)
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
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
#!/usr/bin/env python
import sys, os, re, shutil
import subprocess, shlex, glob
import argparse
import logging
import random, time
import gzip, itertools
try:
  import cdec.score
except ImportError:
  sys.stderr.write('Could not import pycdec, see cdec/python/README.md for details\n')
  sys.exit(1)
have_mpl = True
try: 
  import matplotlib
  matplotlib.use('Agg')
  import matplotlib.pyplot as plt
except ImportError:
  have_mpl = False

#mira run script
#requires pycdec to be built, since it is used for scoring hypothesis
#translations.
#matplotlib must be installed for graphing to work
#email option requires mail

#scoring function using pycdec scoring
def fast_score(hyps, refs, metric):
  scorer = cdec.score.Scorer(metric)
  logging.info('loaded {0} references for scoring with {1}'.format(
                len(refs), metric))
  if metric=='BLEU':
    logging.warning('BLEU is ambiguous, assuming IBM_BLEU\n')
    metric = 'IBM_BLEU'
  elif metric=='COMBI':
    logging.warning('COMBI metric is no longer supported, switching to '
                    'COMB:TER=-0.5;BLEU=0.5')
    metric = 'COMB:TER=-0.5;BLEU=0.5'
  stats = sum(scorer(r).evaluate(h) for h,r in itertools.izip(hyps,refs))
  logging.info('Score={} ({})'.format(stats.score, stats.detail))
  return stats.score

#create new parallel input file in output directory in sgml format
def enseg(devfile, newfile, gprefix):
  try:
    dev = open(devfile)
    new = open(newfile, 'w')
  except IOError, msg:
    logging.error('Error opening source file')
    raise

  i = 0
  for line in dev:
    (src, refs) = line.split(' ||| ', 1)
    if re.match('\s*<seg', src):
      if re.search('id="[0-9]+"', src):
        new.write(line)
      else:
        logging.error('When using segments with pre-generated <seg> tags, '
                      'yout must include a zero based id attribute')
        sys.exit()
    else:
      sgml = '<seg id="{0}"'.format(i)
      if gprefix:
        #TODO check if grammar files gzipped or not
        if os.path.exists('{}.{}.gz'.format(gprefix,i)):
          sgml += ' grammar="{0}.{1}.gz"'.format(gprefix,i)
        elif os.path.exists('{}.{}'.format(gprefix,i)):
          sgml += ' grammar="{}.{}"'.format(gprefix,i)
        else:
          logging.error('Could not find grammar files with prefix '
                        '{}\n'.format(gprefix))
          sys.exit()
      sgml += '>{0}</seg> ||| {1}'.format(src, refs)
      new.write(sgml)
    i+=1
  new.close()
  dev.close()
  return i

def main():
  #set logging to write all info messages to stderr
  logging.basicConfig(level=logging.INFO)
  script_dir = os.path.dirname(os.path.abspath(sys.argv[0]))
  if not have_mpl:
    logging.warning('Failed to import matplotlib, graphs will not be generated.')

  parser= argparse.ArgumentParser(
            formatter_class=argparse.ArgumentDefaultsHelpFormatter)
  parser.add_argument('-d', '--devset', required=True,
                      help='dev set input file in parallel. '
                      'format: src ||| ref1 ||| ref2')
  parser.add_argument('-c', '--config', required=True,
                      help='decoder configuration file')
  parser.add_argument('-w','--weights',
                      help='initial weights file')
  parser.add_argument('-j', '--jobs', type=int, default=1,
                      help='number of decoder processes to run in parallel')
  parser.add_argument('-o','--output-dir', metavar='DIR',
                      help='directory for intermediate and output files. '
                      'defaults to mira.(devset name).(time)')
  parser.add_argument('-e', '--email', 
                      help='email address to send result report')
  parser.add_argument('-t', '--test', 
                      help='test set to decode and evaluate')
  parser.add_argument('--test-config', 
                      help='config file for testing. the config file used '
                      'for tuning feature weights will be used by default.')
  parser.add_argument('-m', '--metric', default='ibm_bleu',
                      help='metric to optimize. Example values: '
                           'ibm_bleu, nist_bleu, Koehn_bleu, TER, Combi')
  parser.add_argument('--max-iterations', type=int, default=20, metavar='N',
                      help='maximum number of iterations to run')
  parser.add_argument('--optimizer', type=int, default=2, choices=range(1,6),
                      help='learning method to use for weight update.'
                      ' Choices: 1) SGD, 2) PA MIRA with Selection from Cutting'
                      ' Plane, 3) Cutting Plane MIRA, 4) PA MIRA,'
                      ' 5) nbest MIRA with hope, fear, and model constraints')
  parser.add_argument('--metric-scale', type=int, default=1, metavar='N',
                      help='scale MT loss by this amount when computing'
                      ' hope/fear candidates')
  parser.add_argument('-k', '--kbest-size', type=int, default=250, metavar='N', 
                      help='size of k-best list to extract from forest')
  parser.add_argument('--update-size', type=int, metavar='N', 
                      help='size of k-best list to use for update. defaults to '
                      'equal kbest-size (applies to optimizer 5)')
  parser.add_argument('--step-size', type=float, default=0.01, 
                      help='controls aggresiveness of update')
  parser.add_argument('--hope', type=int, default=1, choices=range(1,3),
                     help='how to select hope candidate. options: '
                     '1) model score - cost, 2) min cost')
  parser.add_argument('--fear', type=int, default=1, choices=range(1,4),
                      help='how to select fear candidate. options: '
                      '1) model score + cost, 2) max cost, 3) max score')
  parser.add_argument('--sent-approx', action='store_true', 
                      help='use smoothed sentence-level MT metric')
  parser.add_argument('--no-pseudo', action='store_true',
                      help="don't use pseudo document to approximate MT metric")
  parser.add_argument('--no-unique', action='store_true',
                      help="don't extract unique k-best from forest")
  parser.add_argument('-g', '--grammar-prefix', metavar='PATH',
                      help='path to sentence specific grammar files')
  parser.add_argument('--pass-suffix', 
                      help='multipass decoding iteration. see documentation '
                           'at www.cdec-decoder.org for more information')
  args = parser.parse_args()

  args.metric = args.metric.upper()

  if not args.update_size:
    args.update_size = args.kbest_size
  
  #TODO fix path to match decode+evaluate (python month 1-12 instead of 0-11)
  #if an output directory isn't specified, create a unique directory name 
  #of the form mira.(devset).YYYYMMDD-HHMMSS
  if not args.output_dir:
    t = time.localtime()
    args.output_dir = 'mira.{0}.{1}{2:02}{3:02}-{4:02}{5:02}{6:02}'.format(
                      os.path.splitext(args.devset)[0], t[0], t[1], t[2],
                      t[3], t[4], t[5])
    
  if not os.path.isabs(args.output_dir):
    args.output_dir = os.path.abspath(args.output_dir)
  if os.path.exists(args.output_dir):
    if len(os.listdir(args.output_dir))>2:
      logging.error('Error: working directory {0} already exists\n'.format(
                    args.output_dir))
      sys.exit()
  else:
    os.mkdir(args.output_dir)

  if args.grammar_prefix:
    if not os.path.isabs(args.grammar_prefix):
      args.grammar_prefix = os.path.abspath(args.grammar_prefix)
  
  script = open(args.output_dir+'/rerun_mira.sh','w')
  script.write('cd {0}\n'.format(os.getcwd()))
  script.write(' '.join(sys.argv)+'\n')
  script.close()

  #create weights.0 file from initial weights file
  if args.weights:
    shutil.copy(args.weights,os.path.join(args.output_dir,'weights.0'))
  else: #if no weights given, use Glue 0 as default
    weights = open(args.output_dir+'/weights.0','w')
    weights.write('Glue 0\n')
    weights.close()
    args.weights = args.output_dir+'/weights.0'
  
  #create mira ini file
  shutil.copy(args.config,'{0}/kbest_cut_mira.ini'.format(args.output_dir))
  
  newdev = args.output_dir+'/dev.input'
  dev_size = enseg(args.devset, newdev, args.grammar_prefix)
  args.devset = newdev
  
  log_config(args)
  args.weights, hope_best_fear = optimize(args, script_dir, dev_size)
  
  graph_file = ''
  if have_mpl: graph_file = graph(args.output_dir, hope_best_fear, args.metric)

  dev_results, dev_bleu = evaluate(args.devset, args.weights, args.config, 
                         script_dir, args.output_dir)
  if args.test:
    if args.test_config:
      test_results, test_bleu = evaluate(args.test, args.weights, 
                              args.test_config, script_dir, args.output_dir)
    else:
      test_results, test_bleu = evaluate(args.test, args.weights, args.config,
                              script_dir, args.output_dir)
  else: 
    test_results = ''
    test_bleu = ''
  logging.info(dev_results+'\n')
  logging.info(test_results)

  write_report(graph_file, dev_results, dev_bleu, test_results, test_bleu, args)

  if graph_file:
    logging.info('A graph of the best/hope/fear scores over the iterations '
                 'has been saved to {}'.format(graph_file))

  print 'final weights:\n{}\n'.format(args.weights)

#graph of hope/best/fear metric values across all iterations
def graph(output_dir, hope_best_fear, metric):
  max_y = float(max(hope_best_fear['best']))*1.5
  plt.plot(hope_best_fear['best'], label='best')
  plt.plot(hope_best_fear['hope'], label='hope')
  plt.plot(hope_best_fear['fear'], label='fear')
  plt.axis([0,len(hope_best_fear['fear'])-1,0,max_y])
  plt.xlabel('Iteration')
  plt.ylabel(metric)
  plt.legend()
  graph_file = output_dir+'/mira.pdf'
  plt.savefig(graph_file)
  return graph_file

#evaluate a given test set using decode-and-evaluate.pl
def evaluate(testset, weights, ini, script_dir, out_dir):
  evaluator = '{}/../utils/decode-and-evaluate.pl'.format(script_dir)
  try:
    p = subprocess.Popen([evaluator, '-c', ini, '-w', weights, '-i', testset, 
                         '-d', out_dir], stdout=subprocess.PIPE)
    results, err = p.communicate()
    bleu, results = results.split('\n',1)
  except subprocess.CalledProcessError:
    logging.error('Evalutation of {} failed'.format(testset))
    results = ''
    bleu = ''
  return results, bleu

#print a report to out_dir/mira.results
#send email with results if email was given
def write_report(graph_file, dev_results, dev_bleu, 
                 test_results, test_bleu, args):
  features, top, bottom = weight_stats(args.weights) 
  top = [f+' '+str(w) for f,w in top]
  bottom = [f+' '+str(w) for f,w in bottom]
  subject = 'MIRA {0} {1:7}'.format(os.path.basename(args.devset), dev_bleu)
  if args.test:
    subject += ' {0} {1:7}'.format(os.path.basename(args.test), test_bleu)

  message = ('MIRA has finished running. '+
            'The final weights can be found at \n{}\n'.format(args.weights)+
            'Average weights across all iterations '+
            '\n{}/weights.average\n'.format(args.output_dir)+
            'Weights were calculated for {} features\n\n'.format(features)+
            '5 highest weights:\n{}\n\n'.format('\n'.join(top))+
            '5 lowest weights:\n{}\n'.format('\n'.join(bottom)))
  
  if dev_results:
    message += '\nEvaluation: dev set\n{}'.format(dev_results)
  if test_results:
    message += '\nEvaluation: test set\n{}'.format(test_results)
 
  out = open(args.output_dir+'/mira.results','w')
  out.write(message)
  out.close()
 
  if args.email:
    cmd = ['mail', '-s', subject]
    if graph_file:
      cmd += ['-a', graph_file]
    email_process = subprocess.Popen(cmd+[args.email], stdin = subprocess.PIPE)
    email_process.communicate(message)

#feature weights stats for report
def weight_stats(weight_file):
  f = open(weight_file)
  features = []
  for line in f:
    feat, weight = line.strip().split()
    features.append((feat,float(weight)))
  features.sort(key=lambda a: a[1], reverse=True)
  return len(features), features[:5], features[-5:]

#create source and refs files from parallel devset
#TODO remove when kbest_cut_mira changed to take parallel input
def split_devset(dev, outdir):
  parallel = open(dev)
  source = open(outdir+'/source.input','w')
  refs = open(outdir+'/refs.input', 'w')
  references = []
  for line in parallel:
    s,r = line.strip().split(' ||| ',1)
    source.write(s+'\n')
    refs.write(r+'\n')
    references.append(r)
  source.close()
  refs.close()
  return (outdir+'/source.input', outdir+'/refs.input')

def optimize(args, script_dir, dev_size):
  parallelize = script_dir+'/../utils/parallelize.pl'
  decoder = script_dir+'/kbest_cut_mira'
  (source, refs) = split_devset(args.devset, args.output_dir)
  port = random.randint(15000,50000)
  logging.info('using port {}'.format(port))
  num_features = 0
  last_p_score = 0
  best_score_iter = -1
  best_score = -1
  i = 0
  hope_best_fear = {'hope':[],'best':[],'fear':[]}
  #main optimization loop
  while i<args.max_iterations:
    logging.info('======= STARTING ITERATION {} ======='.format(i))
    logging.info('Starting at {}'.format(time.asctime()))

    #iteration specific files
    runfile = args.output_dir+'/run.raw.'+str(i)
    onebestfile = args.output_dir+'/1best.'+str(i)
    logdir = args.output_dir+'/logs.'+str(i)
    decoderlog = logdir+'/decoder.sentserver.log.'+str(i)
    weightdir = args.output_dir+'/weights.pass'+str(i)
    os.mkdir(logdir)
    os.mkdir(weightdir)
    weightsfile = args.output_dir+'/weights.'+str(i)
    logging.info('  log directory={}'.format(logdir))
    curr_pass = '0{}'.format(i)
    decoder_cmd = ('{0} -c {1} -w {2} -r{3} -m {4} -s {5} -b {6} -k {7} -o {8}'
                   ' -p {9} -O {10} -D {11} -h {12} -f {13} -C {14}').format(
                   decoder, args.config, weightsfile, refs, args.metric,
                   args.metric_scale, args.update_size, args.kbest_size, 
                   args.optimizer, curr_pass, weightdir, args.output_dir,
                   args.hope, args.fear, args.step_size)
    if not args.no_unique: 
      decoder_cmd += ' -u'
    if args.sent_approx:
      decoder_cmd += ' -a'
    if not args.no_pseudo:
      decoder_cmd += ' -e'
    
    #always use fork 
    parallel_cmd = '{0} --use-fork -e {1} -j {2} --'.format(
                    parallelize, logdir, args.jobs)
    
    cmd = parallel_cmd + ' ' + decoder_cmd
    logging.info('OPTIMIZATION COMMAND: {}'.format(cmd))
   
    dlog = open(decoderlog,'w')
    runf = open(runfile,'w')
    retries = 0
    num_topbest = 0

    while retries < 6:
      #call decoder through parallelize.pl
      p1 = subprocess.Popen(['cat', source], stdout=subprocess.PIPE)
      exit_code = subprocess.call(shlex.split(cmd), stderr=dlog, stdout=runf, 
                                  stdin=p1.stdout)
      p1.stdout.close()
      
      if exit_code:
        logging.error('Failed with exit code {}'.format(exit_code))
        sys.exit(exit_code)

      try:
        f = open(runfile)
      except IOError, msg:
        logging.error('Unable to open {}'.format(runfile))
        sys.exit()
      
      num_topbest = sum(1 for line in f)
      f.close()
      if num_topbest == dev_size: break
      logging.warning('Incorrect number of top best. Sleeping for 10 seconds and retrying...')
      time.sleep(10)
      retries += 1
    
    if dev_size != num_topbest:
      logging.error("Dev set contains "+dev_size+" sentences, but we don't "
                    "have topbest for all of these. Decoder failure? "
                    " Check "+decoderlog)
      sys.exit()
    dlog.close()
    runf.close()

    #write best, hope, and fear translations
    run = open(runfile)
    H = open(runfile+'.H', 'w')
    B = open(runfile+'.B', 'w')
    F = open(runfile+'.F', 'w')
    hopes = []
    bests = []
    fears = []
    for line in run:
      hope, best, fear = line.split(' ||| ')
      hopes.append(hope)
      bests.append(best)
      fears.append(fear)
      H.write('{}\n'.format(hope))
      B.write('{}\n'.format(best))
      F.write('{}\n'.format(fear))
    run.close()
    H.close()
    B.close()
    F.close()

    #gzip runfiles and log files to save space
    gzip_file(runfile)
    gzip_file(decoderlog)

    ref_file = open(refs)
    references = [line.split(' ||| ') for line in 
                  ref_file.read().strip().split('\n')]
    ref_file.close()
    #get score for best hypothesis translations, hope and fear translations
    dec_score = fast_score(bests, references, args.metric)
    dec_score_h = fast_score(hopes, references, args.metric)
    dec_score_f = fast_score(fears, references, args.metric)
    
    hope_best_fear['hope'].append(dec_score)
    hope_best_fear['best'].append(dec_score_h)
    hope_best_fear['fear'].append(dec_score_f)
    logging.info('DECODER SCORE: {0} HOPE: {1} FEAR: {2}'.format(
                  dec_score, dec_score_h, dec_score_f))
    if dec_score > best_score:
      best_score_iter = i
      best_score = dec_score

    new_weights_file = '{}/weights.{}'.format(args.output_dir, i+1)
    last_weights_file = '{}/weights.{}'.format(args.output_dir, i)
    i += 1
    weight_files = weightdir+'/weights.mira-pass*.*[0-9].gz'
    average_weights(new_weights_file, weight_files)

  logging.info('BEST ITERATION: {} (SCORE={})'.format(
               best_score_iter, best_score))
  weights_final = args.output_dir+'/weights.final'
  logging.info('WEIGHTS FILE: {}'.format(weights_final))
  shutil.copy(last_weights_file, weights_final)
  average_final_weights(args.output_dir)
  
  return weights_final, hope_best_fear

#TODO
#create a weights file with the average of the weights from each iteration
def average_final_weights(out_dir):
  logging.info('Average of weights from each iteration\n')
  weight_files = glob.glob(out_dir+'/weights.[1-9]*')
  features = {}
  for path in weight_files:
    weights = open(path)
    for line in weights:
      f, w = line.strip().split(' ', 1)
      if f in features:
        features[f] += float(w)
      else:
        features[f] = float(w)
    weights.close()

  out = open(out_dir+'/weights.average','w')
  for f in iter(features):
    out.write('{} {}\n'.format(f,features[f]/len(weight_files)))
  logging.info('An average weights file can be found at' 
               '\n{}\n'.format(out_dir+'/weights.average'))

#create gzipped version of given file with name filename.gz
# and delete original file
def gzip_file(filename):
  gzip_file = gzip.open(filename+'.gz','wb')
  f = open(filename)
  gzip_file.writelines(f)
  f.close()
  gzip_file.close()
  os.remove(filename)

#average the weights for a given pass
def average_weights(new_weights, weight_files):
  logging.info('AVERAGE {} {}'.format(new_weights, weight_files))
  feature_weights = {}
  total_mult = 0.0
  for path in glob.glob(weight_files):
    score = gzip.open(path)
    mult = 0
    logging.info('  FILE {}'.format(path))
    msg, ran, mult = score.readline().strip().split(' ||| ')
    logging.info('  Processing {} {}'.format(ran, mult))
    for line in score:
      f,w = line.split(' ',1)
      if f in feature_weights:
        feature_weights[f]+= float(mult)*float(w)
      else: 
        feature_weights[f] = float(mult)*float(w)
    total_mult += float(mult)
    score.close()
  
  #write new weights to outfile
  logging.info('Writing averaged weights to {}'.format(new_weights))
  out = open(new_weights, 'w')
  for f in iter(feature_weights):
    avg = feature_weights[f]/total_mult
    out.write('{} {}\n'.format(f,avg))

def log_config(args):
  logging.info('WORKING DIRECTORY={}'.format(args.output_dir))
  logging.info('INI FILE={}'.format(args.config))
  logging.info('DEVSET={}'.format(args.devset))
  logging.info('EVAL METRIC={}'.format(args.metric))
  logging.info('MAX ITERATIONS={}'.format(args.max_iterations))
  logging.info('PARALLEL JOBS={}'.format(args.jobs))
  logging.info('INITIAL WEIGHTS={}'.format(args.weights))
  if args.grammar_prefix:
    logging.info('GRAMMAR PREFIX={}'.format(args.grammar_prefix))
  if args.test:
    logging.info('TEST SET={}'.format(args.test))
  else:
    logging.info('TEST SET=none specified')
  if args.test_config:
    logging.info('TEST CONFIG={}'.format(args.test_config))
  if args.email:
    logging.info('EMAIL={}'.format(args.email))

if __name__=='__main__':
  main()