forked from maximilianh/pubMunch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpubClassify
executable file
·1199 lines (1055 loc) · 47.4 KB
/
pubClassify
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
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#!/usr/bin/env python
# script to run the various steps of the text classification pipeline
# first load the standard libraries from python
# we require at least python 2.5
#from sys import *
import sys
if sys.version_info[0]==2 and not sys.version_info[1]>=7:
print "Sorry, this program requires at least python 2.7"
sys.exit(1)
# load default python packages
import logging, optparse, os, glob, zipfile, types, gzip, shutil, subprocess, \
itertools, operator, gc, random, codecs
import marshal
from os.path import *
from collections import defaultdict, Counter
from datetime import datetime
# add <scriptDir>/lib/ to package search path
progFile = os.path.abspath(sys.argv[0])
progDir = os.path.dirname(progFile)
pubToolsLibDir = os.path.join(progDir, "lib")
sys.path.insert(0, pubToolsLibDir)
# now load our own libraries
import pubGeneric, maxRun, pubConf, maxCommon, pubExpMatrix, html, pubStore, pubAlg
# === CONSTANTS & GLOBALS ===================================
# do not delete output dirs when starting a new step
leaveDirs = False
# === COMMAND LINE INTERFACE, OPTIONS AND HELP ===
parser = optparse.OptionParser("""usage: %prog [options] <datasetList> <step> - script to run the various steps of the article classification pipeline
steps:
bestWords = create the list of best words for each database
wordCount = create document-frequency list of words, needs to be filtered with
google/BNC to be useful
writes to data/wordList/wordList.raw.txt
tmatrix = create training matrix from document collection (map/reduce)
reads pmid lists from data/classify/pmids
writes to <pubBase>/classify/tmatrix.tab
rewrite training matrix to one file per biological database
and train svmlight models
writes to data/classify/models
dmatrix = create big document matrix from document collection, by default
runs only on
reads from article datasets
writes to <pubBase>/classify/docMatrix.svml
classify = run models onto docMatrix and write to <pubBase>/classify/docClasses.tab
html = generate html pages for databases and write to html directory
example:
pubFilter pmc,elsevier,crawler dmatrix
""")
parser.add_option("", "--skipMap", dest="skipMap", action="store_true", help="skip all map steps")
parser.add_option("", "--onlyDbs", dest="onlyDbs", action="store", help="run only on a given db, not on all")
parser.add_option("-t", "--test", dest="test", action="store_true", help="only run the test, nothing else")
parser.add_option("-g", "--twoGrams", dest="twoGrams", action="store_true", help="also use twograms", default=False)
parser.add_option("-w", "--bestWordCount", dest="bestWordCount", action="store", type="int", help="how many best words to use? default %default", default=2000)
parser.add_option("", "--noMeta", dest="noMeta", action="store_true", help="do not add separate features for title/abstract/lastAuthor", default=False)
parser.add_option("-l", "--leaveDirs", dest="leaveDirs", action="store_true", help="do not delete output dirs before starting a new step, saves times but can corrupt data", default=False)
parser = pubGeneric.addGeneralOptions(parser)
(options, args) = parser.parse_args()
pubGeneric.setupLogging(__file__, options)
# ==== FILE NAMES ====
# set up two main dirs: one for static data from UCSC and one for local data
dataDir = pubConf.getStaticDataDir()
statClassDir = join(dataDir, "classify") # static data, part of code repo: foreground PMIDs, one list per DB
varDir = join(pubConf.pubsDataDir, "classify") # pipeline data, like temporary files and output
localDir = join(pubConf.localHeadDir, "classify")
rawWordListFname = join(statClassDir, "wordList.raw.txt")
wordListFname = join(statClassDir, "wordList.txt.gz")
pmidListDir = join(statClassDir, "trainPmids")
# like above, but ALL pmids in database
pmidListDirOrig = join(statClassDir, "trainPmids/orig")
svmlBinDir = pubConf.svmlBinDir
def defDirectories(baseDir, locBaseDir, datasets, options):
class D:
pass
maxWordRank = options.bestWordCount
if not isdir(baseDir):
os.makedirs(baseDir)
gramType = "1grams"
if options.twoGrams:
gramType = "2grams"
sepType = ".withMeta"
if options.noMeta:
sepType = ""
dirs = D()
dirs.wordCountDir = join(baseDir, "wordListOutput%s.%s" % (sepType, gramType))
#dirs.concatDir = join(baseDir, "wordListByDb")
dirs.concatDir = join(locBaseDir, "wordListByDb%s.%s" % (sepType, gramType))
dirs.corpusDir = join(locBaseDir, "corpus")
dirs.wordCountFname = join(baseDir, "wordCounts.marshal")
dirs.wordPmidDir = join(baseDir, "wordPmidDicts.%s" % (gramType))
dirs.textDirs = pubConf.resolveTextDirs(datasets)
dirs.rankedWordsDir = join(baseDir, "rankedWords.%s.%d" % (gramType, maxWordRank))
dirs.bestWordsDir = join(baseDir, "bestWords.%s.%d" % (gramType, maxWordRank))
dirs.tMatrixFname = join(baseDir, "allTrain.pmidsvml")
dirs.clusterSvmlDir = join(baseDir, "svmlJobOutput.%s.%d" % (gramType, maxWordRank))
dirs.svmlDir = join(baseDir, "svmlInput.%s.%d" % (gramType, maxWordRank))
dirs.trainSvmlDir = join(baseDir, "svmlTrain.%s.%d" % (gramType, maxWordRank))
dirs.testSvmlDir = join(baseDir, "svmlTest.%s.%d" % (gramType, maxWordRank))
dirs.benchTable = join(baseDir, "bench.%s.%d.tab" % (gramType, maxWordRank))
dirs.modelDir = join(baseDir, "svmlModels.%s.%d" % (gramType, maxWordRank))
dirs.alphaDir = join(baseDir, "svmlAlphas.%s.%d" % (gramType, maxWordRank))
dirs.dMatrixFname = join(baseDir, "docs.svml")
dirs.dIdFname = join(baseDir, "docs.docIds")
dirs.categoryFname = join(baseDir, "docClasses.tab")
dirs.predClassDir = dirs.testSvmlDir
return dirs
# ==== FUNCTIONs =====
def parsePmidsDbs(inDir):
" return training PMIDs in format pmid -> list of DBs "
pmids, dbList, dbPmids = parsePmids(inDir)
res = defaultdict(list)
for pmid, classDbs in dbPmids.iteritems():
dbList = []
for docClass, db in classDbs:
dbList.append(db)
res[pmid]=dbList
return res
def parsePmidClasses(inDir, db):
""" return training PMIDs in format pmid -> class (1 or -1)"""
logging.info("Parsing class assignments for db %s" % db)
overlapPmids = set()
res = {}
for className, classId in [("pos", 1), ("neg", -1)]:
inPmids = set(open(join(inDir, className+"."+db+".txt")).read().splitlines())
for pmid in inPmids:
pmid = int(pmid)
if pmid in res:
#raise Exception("PMID %d is in both classes" % pmid)
overlapPmids.add(pmid)
res[pmid] = classId
logging.warn("%d PMIDs in both classes" % len(overlapPmids))
return res
def parsePosPmids(db, inDir):
" parse pos pmids of db"
fname = join(inDir, "pos.%s.txt" % db)
pmids = open(fname).read().splitlines()
pmids = [int(x) for x in pmids]
return set(pmids)
def parsePmids(inDir):
""" parse pmids from dir and return as
- a set of all PMIDs,
- a list of DBS and -
- as a dict pmid -> list of (class, db) tuples, class is either "+1" or "-1"
all PMIDs are integers
"""
logging.info("Parsing PMIDs")
allPmids = set()
pmidDbs = {}
dbs = set()
inFnames = glob.glob(join(inDir, "*.txt"))
for inFname in inFnames:
base = basename(inFname)
if not (base.startswith("pos") or base.startswith("neg")):
continue
pmidClass, db = base.split(".")[:2]
dbs.add(db)
if pmidClass=="pos":
svmlClass = "+1"
elif pmidClass=="neg":
svmlClass = "-1"
else:
assert(False)
logging.debug("Parsing %s" % inFname)
for line in open(inFname):
pmid = line.strip()
pmid = int(pmid)
allPmids.add(pmid)
pmidDbs.setdefault(pmid, []).append( (svmlClass, db) )
logging.info("Read %d PMIDs from %s" % (len(allPmids), inDir))
#for pmid, in pmidDbs.iteritems():
#logging.info("Got %d pmids for DB %s" % (len(pmids), db))
assert(len(pmidDbs)!=0)
assert(len(allPmids)!=0)
return allPmids, dbs, pmidDbs
def mkEmptyDir(dir, doNotDelete=False):
" make sure dir exists and is empty "
logging.info("Making/Cleaning %s" % dir)
if leaveDirs:
doNotDelete=True
if isdir(dir):
if not doNotDelete:
shutil.rmtree(dir)
#cmd = "mv %(dir)s %(dir)s.old; rm -rf %(dir)s.old &" % locals()
#print cmd
#os.system(cmd)
if not isdir(dir):
os.makedirs(dir)
def splitSvml(tMatrixFname, dbs, pmidDbs, svmlDir):
""" create one svml output file in svmlDir per db in dbPmids and distribute the svml lines from
tMatrixFname to the right files in svmlDir
"""
dbOfh = {}
fnames = []
mkEmptyDir(svmlDir)
for db in dbs:
ofname = join(svmlDir, db+".svml")
dbOfh[db] = open(ofname, "w")
fnames.append(ofname)
logging.debug("Rewriting %s" % tMatrixFname)
for line in open(tMatrixFname):
docId, featVec = line.split(" ", 1)
pmid = int(docId.split("/")[2])
for svmlClass, db in pmidDbs[pmid]:
dbOfh[db].write(svmlClass+" ")
dbOfh[db].write(featVec)
logging.info("Wrote SVML files: %s" % " ".join(fnames))
def svmlLearn(svmlBinDir, svmlDir, modelDir, alphaDir, dbList):
" run svml_learn on all .svml files in svmlDir "
binPath = join(svmlBinDir, "svm_learn")
if not isfile(binPath):
raise Exception("%s does not exist" % binPath)
mkEmptyDir(modelDir)
mkEmptyDir(alphaDir)
logging.info("Using SVML files in dir %s" % svmlDir)
for svmlFname in glob.glob(join(svmlDir, "*.svml")):
db = splitext(basename(svmlFname))[0]
if dbList!=None and db not in dbList:
continue
modelFname = join(modelDir, db+".model")
alphaFname = join(alphaDir, db+".alpha")
logging.info("Running SVMlight for db %s" % db)
cmd = [binPath, svmlFname, modelFname, "-a", alphaFname]
subprocess.check_call(cmd)
logging.info("alphaput written to %s and %s" % (modelDir, alphaDir))
def svmlClassify(svmlBinDir, svmlDir, modelDir, classDir, dbList):
" run svml on all models from modelDir "
#mkEmptyDir(classDir)
binPath = join(svmlBinDir, "svm_classify")
if not isfile(binPath):
raise Exception("%s does not exist" % binPath)
runner = pubGeneric.makeClusterRunner(__file__, algName="svmlClassify", headNode='localhost')
for modelFname in glob.glob(join(modelDir, "*.model")):
db = splitext(basename(modelFname))[0]
if dbList!=None and db not in dbList:
continue
svmlFname = join(svmlDir, db+".svml")
#logging.info("Classifying with SVMLight, feature file %s" % svmlFname)
outFname = join(classDir, db+".predClasses")
logging.info("Running on %s and %s" % (svmlFname, modelFname))
cmd = [binPath, svmlFname, modelFname, "{check out line+ %s}" % outFname]
cmd = " ".join(cmd)
#logging.debug("command is %s" % cmd)
runner.submit(cmd)
runner.finish(wait=True)
def parseDocClasses(fname):
" return dict db -> list of (articleId (INT!), score) "
logging.info("Parsing %s" % fname)
res = defaultdict(list)
for row in maxCommon.iterTsvRows(fname):
artId = int(row.articleId)
classes = row.classes.split(',')
scores = row.scores.split(',')
for db, score in zip(classes, scores):
res[db].append((artId, float(score)))
return res
def makeHtmlTest(dbList, datasets, svmlDir, outDir):
" "
mkEmptyDir(outDir)
logging.info("Writing html to %s" % outDir)
for db in dbList:
pmidFn = join(svmlDir, db+".pmids")
pmids = open(pmidFn).read().splitlines()
predFn = join(svmlDir, db+".predClasses")
predScores = open(predFn).read().splitlines()
realFn = join(svmlDir, db+".classes")
realClasses = open(realFn).read().splitlines()
logging.info("Reading %s, %s and %s" % (pmidFn, predFn, realFn))
outfname = join(outDir, db+".html")
errDict = defaultdict(list)
for pmid, predScore, realClass in zip(pmids, predScores, realClasses):
predScore = float(predScore)
realClass = int(realClass)
if predScore>0 and realClass==1:
err = "tp"
elif predScore<0 and realClass==-1:
err = "tn"
elif predScore>0 and realClass==-1:
err = "fp"
elif predScore<0 and realClass==1:
err = "fn"
errDict[err].append((pmid, predScore, realClass))
logging.info("DB %s, filename %s" % (db, outfname))
h = html.htmlWriter(outfname)
desc = {"tp" : "in database and predicted", \
"tn" : "not in database and predicted", \
"fp" : "not in database, but predicted to be", \
"fn" : "in database, but not predicted"
}
h.head("Database: %s" % db)
h.h4("Database: %s" % db)
h.p()
h.writeLn("<ul>")
for errCode, pmidList in errDict.iteritems():
h.li("Class: %s (%s): %d documents" % (errCode, desc[errCode], len(pmidList)))
h.writeLn("</ul>")
#h.hr()
for errType, errTuples in errDict.iteritems():
if errType in ["tp", "tn"]:
continue
h.h4("Class: %s / %s" % (errType, desc[errType]))
#pm = maxCommon.ProgressMeter(pmidCount, stepCount=100)
for pmid, predScore, realClass in errTuples:
art = pubStore.lookupArticleByPmid(datasets, pmid)
if art==None:
logging.warn("No info on PMID %s?" % pmid)
continue
h.link(art["fulltextUrl"], art["title"])
h.br()
ref = ["score: "+str(predScore), art["journal"], art["year"], art["authors"]]
h.writeLn(", ".join(ref))
#h.hr()
h.p()
#pm.taskCompleted()
h.endHtml()
def readWordPmids3(inDir):
" merge together all those word -> pmid dictionatries and return it as one big dict "
gc.disable()
fnames = glob.glob(join(inDir, "*.marshal"))
logging.info("Found %d files in %s" % (len(fnames), inDir))
duplPmidCount = 0
donePmids = set()
wordPmids = {}
pm = maxCommon.ProgressMeter(len(fnames))
for fname in fnames:
logging.debug("Reading %s" % fname)
if os.path.getsize(fname)==0:
logging.error("Empty file %s, skipping" % fname)
continue
f = open(fname)
fileWordPmids = marshal.load(f)
f.close()
for word, pmids in fileWordPmids.iteritems():
wordPmids.setdefault(word, []).extend(pmids)
pm.taskCompleted()
logging.info("wordCount: %d" % len(wordPmids))
gc.enable()
return wordPmids
def readWordPmids(inDir, pmids, onlyOneGrams=False):
" parse wordCounts, return as dict word -> pmids "
#fnames = glob.glob(join(inDir, "*.tab.gz"))
logging.info("Found %d files in %s" % (len(fnames), inDir))
logging.info("Reading wordcounts for %d PMIDs" % len(pmids))
logging.info("Time: %s" % str(datetime.now()))
pm = maxCommon.ProgressMeter(len(fnames))
donePmids = set()
duplPmidCount = 0
res = {}
for fname in fnames:
logging.debug("Parsing %s")
for row in maxCommon.iterTsvRows(fname):
gc.disable()
pmid = int(row.pmid)
if pmid not in pmids:
logging.debug("Ignoring %s, not in target pmids" % pmid)
continue
if pmid in donePmids:
logging.debug("duplicated pmid %d" % pmid)
duplPmidCount += 1
continue
donePmids.add(pmid)
wordStr = row.wordCounts
words = [w.split("=")[0] for w in wordStr.split(",")]
for word in words:
if onlyOneGrams and "_" in word:
continue
word = str(word) # bytestrings take 1/2 memory
res.setdefault(word, []).append(pmid)
gc.enable()
pm.taskCompleted()
logging.info("Featurecount: %d" % len(res))
logging.info("input PMID count: %d" % len(pmids))
logging.info("found PMID count: %d" % len(donePmids))
logging.info("Duplicated PMIDs: %d" % duplPmidCount)
return res, donePmids
def filterPmids(inDir, pmids, dbList, outDir):
" parse all wordCounts, write to one single file per DB "
maxCommon.mustExistDir(outDir, makeDir=True)
fnames = glob.glob(join(inDir, "*.tab.gz"))
logging.info("Found %d files in %s" % (len(fnames), inDir))
logging.info("Reading wordcounts for %d PMIDs" % len(pmids))
logging.info("Time: %s" % str(datetime.now()))
pm = maxCommon.ProgressMeter(len(fnames))
# open filehandles
headerLine = gzip.open(fnames[0]).readline()
ofhs = {}
for db in dbList:
ofhs[db] = gzip.open(join(outDir, db+".tab.gz"), "w")
ofhs[db].write(headerLine)
# go over input files and write lines
donePmids = set()
duplPmidCount = 0
res = {}
for fname in fnames:
logging.debug("Parsing %s")
for row, line in maxCommon.fastIterTsvRows(fname):
pmid = int(row.pmid)
if pmid not in pmids:
logging.debug("Ignoring %s, not in target pmids" % pmid)
continue
if pmid in donePmids:
logging.debug("duplicated pmid %d" % pmid)
duplPmidCount += 1
continue
donePmids.add(pmid)
dbList = pmids[pmid]
for db in dbList:
ofhs[db].write(line)
ofhs[db].write("\n")
pm.taskCompleted()
logging.info("input PMID count: %d" % len(pmids))
logging.info("found PMID count: %d" % len(donePmids))
logging.info("Duplicated PMIDs: %d" % duplPmidCount)
logging.info("Output directory: %s" % outDir)
def calcChiSq(allCount, posCount, negCount, posOvl, negOvl):
expOvl = int((float(negOvl) / negCount) * posCount)
if expOvl==0:
expOvl=1
chiSq = (posOvl-expOvl)**2 / float(expOvl)
return chiSq
def makeBestWords(dbList, wordCountFname, pmidDir, rankedWordsDir, maxRank, bestWordsDir):
""" read word counts from dataset completely into memory, for each db, use
pmidList to separate into pos/neg and do a chi-square test to find best
words
"""
mkEmptyDir(bestWordsDir)
mkEmptyDir(rankedWordsDir)
logging.info("Loading word->doc table %s" % wordCountFname)
wordPmids = marshal.load(open(wordCountFname))
for db in dbList:
logging.info("Finding best words for db %s" % db)
# parse pmids
posPmids = set([int(l.strip()) for l in open(join(pmidDir, "pos."+db+".txt"))])
negPmids = set([int(l.strip()) for l in open(join(pmidDir, "neg."+db+".txt"))])
logging.debug("%d positive documents, %d background documents" % (len(posPmids), len(negPmids)))
gc.disable()
wordScores = []
for word, pmidList in wordPmids.iteritems():
wordPmids = set(pmidList)
obsOvl = len(wordPmids.intersection(posPmids))
if obsOvl < 10:
continue
negOvl = len(wordPmids.intersection(negPmids))
expOvl = int(float(negOvl) * len(posPmids))
logging.debug("word %s: overlap with positives %d, overlap with background %d, expected overlap with positives %d" % (word, obsOvl, negOvl, expOvl))
if expOvl==0:
expOvl=1
chiSq = (obsOvl-expOvl)**2 / float(expOvl)
wordScores.append( (word, chiSq) )
wordScores.sort(key=operator.itemgetter(1))
gc.enable()
# write words + score
ofname = join(rankedWordsDir, "%s.tab" % (db))
ofh = open(ofname, "w")
ofh.write("word\tchi2\n")
for word, score in wordScores:
ofh.write("%s\t%f\n" % (word, score))
logging.info("Wrote chi2 scores to %s" % ofname)
# write only top x words
ofname = join(bestWordsDir, "%s.txt" % (db))
ofh = open(ofname, "w")
for word, score in wordScores[:maxRank]:
ofh.write("%s\n" % (word))
logging.info("Wrote top %d words to %s" % (maxRank, ofname))
def filterWords(wordCounts, posMin, posMax, negMin, negMax):
logging.info("Basic filter: Keeping only words with count: %d<pos<%d, %d<neg<%d" % (posMin, posMax, negMin, negMax))
#logging.info("Time: %s" % str(datetime.now()))
gc.disable()
newDict = {}
for key, valList in wordCounts.iteritems():
count = len(valList)
if count>posMin and count<posMax and \
count>negMin and count<negMax:
newDict[key] = valList
gc.enable()
logging.info("Word count reduced from %d to %d after basic filtering" % (len(wordCounts), len(newDict)))
return newDict
def writeWordPmids(wordPmids, wordCountFname):
logging.info("Writing words")
f = open(wordCountFname, "w")
marshal.dump(wordPmids, f)
f.close()
logging.info("%s: %3d MB" % (wordCountFname, os.path.getsize(wordCountFname)/1000000))
def getDbList(pmidDir):
" create a list of all possible DBs in pmidDir, those with a pos. and a neg.<db>.txt file "
fnames = os.listdir(pmidDir)
posNames = [fn.split(".")[1] for fn in fnames if fn.startswith("pos.") and fn.endswith(".txt")]
negNames = [fn.split(".")[1] for fn in fnames if fn.startswith("neg.") and fn.endswith(".txt")]
dbNames = set(posNames).intersection(negNames)
return dbNames
def submitToSvmlJobs(textDirs, outDir, wordFname, addMeta, addTwoGrams):
paramDict = { "dbWords" : dbWords, "addMeta":addMeta, "addTwoGrams" : addTwoGrams }
algName = "wordList.py:SvmlWriter"
runner = pubGeneric.makeClusterRunner(__file__, algName="toSvml")
pubAlg.submitAnnotateWrite("wordList", textDirs, paramDict, outDir, runner=runner)
runner.finish()
def submitWordCountJobs(pmids, datasets, textDirs, outDir, addMeta, addTwoGrams):
"""
creates a big marshalled dict with word -> list of pmids
"""
maxCommon.mustExistDir(outDir, makeDir=True)
mkEmptyDir(outDir)
runner = pubGeneric.makeClusterRunner(__file__, algName="countWords")
paramDict = {"addMeta" : addMeta, "addTwoGrams" : addTwoGrams, "pmids":pmids}
pubAlg.submitAnnotateWrite(runner, "wordList", textDirs, paramDict, outDir)
logging.info("Output written to %s" % outDir)
runner.finish()
def submitWordCountJobs2(pmidListDir, datasets, textDirs, outDir, addMeta, addTwoGrams):
"""
creates a big marshalled dict with word -> list of pmids
"""
#maxCommon.mustExistDir(outDir, makeDir=True)
#mkEmptyDir(outDir)
pmids, dbs, dbPmids = parsePmids(pmidListDir)
runner = pubGeneric.makeClusterRunner(__file__, algName="countWords")
paramDict = {
"addMeta" : addMeta,
"addTwoGrams" : addTwoGrams,
"outFname" : "big.marshal",
"pmids" : pmids
}
pubAlg.mapReduce("wordList:WordPmids", textDirs, paramDict, None, runner=runner, runTest=True)
logging.info("Output written to %s" % outDir)
runner.finish()
def wordCountToSvml(dbList, pmidListDir, concatDir, bestWordsDir, svmlDir):
" convert our internal word=count format to svmlight "
mkEmptyDir(svmlDir)
for db in dbList:
# setup files
pmidClasses = parsePmidClasses(pmidListDir, db)
#inFname = join(concatDir, db+".tab.gz")
inFname = join(concatDir, db+".tab")
svmlOutFname = join(svmlDir, db+".svml")
pmidOutFname = join(svmlDir, db+".pmids")
logging.info("Writing to %s and %s" % (svmlOutFname, pmidOutFname))
svmlFh = open(svmlOutFname, "w")
pmidFh = open(pmidOutFname, "w")
wordFname = join(bestWordsDir, db+".txt")
logging.info("Reading bestwords from %s" % wordFname)
# setup dict word -> integer
bestWordList = open(wordFname).read().splitlines()
wordToIdx = dict((y,x+1) for x,y in enumerate(bestWordList))
# parse infile and convert to svml
logging.info("Parsing %s" % inFname)
gc.disable()
badCount = 0
rowCount = 0
for row, line in maxCommon.fastIterTsvRows(inFname):
pmid = int(row.pmid)
classId = pmidClasses[pmid]
# create tuples (word, count)
tuples = [tuple(tpl.split("=")) for tpl in row.wordCounts.split(",")]
#if pmid==14593087:
#print list(tuples)
# create list of (word, count) for all good words
items = [(wordToIdx[t[0]], int(t[1])) for t in tuples if t[0] in wordToIdx]
#if pmid==14593087:
#print list(countTuples)
# convert to sorted list (wordIdx, count) and join to string
if len(items)==0:
if classId==1:
logging.warn("PMID %d, count %d: positive class but no single good word?" % \
(badCount, pmid))
badCount += 1
continue
items.sort(key=operator.itemgetter(0))
items = ("%d:%d" % (x,y) for x,y in items)
svmlStr = " ".join(items)
svmlFh.write("%+d %s\n" % (classId, svmlStr))
pmidFh.write("%d\n" % pmid)
rowCount +=1
if rowCount % 1000 == 0:
print rowCount
gc.enable()
def getText(textDirs, dbList, pmidToDb, corpusDir):
mkEmptyDir(corpusDir)
# open outfiles
outFiles = {}
for db in dbList:
outFname = join(corpusDir, db+".txt")
logging.info("Opening %s" % outFname)
outFiles[db] = open(outFname, "w")
# iterate over input and spread over output
logging.debug("Reading text from %s" % textDirs)
for textDir in textDirs:
#ar = pubStore.PubReaderFile(textDir)
print textDir
dataIter = pubStore.iterArticleDirList(textDir, onlyMain=True, preferPdf=True)
#for article, fileList in ar.iterArticlesFileList(onlyMain=True, onlyBestMain=True):
for article, fileList in dataIter:
f = fileList[0]
text = f.content
if article.pmid=="":
continue
pmid = int(article.pmid)
if pmid!="" and pmid in pmidToDb:
for db in pmidToDb[pmid]:
outFiles[db].write("%s\t%s\n" % (pmid, text.encode("utf8")))
def dictToSvml (wordToPmids, bestWords, posPmids, negPmids, svmlDir, db):
pmidToWords = defaultdict(list)
for wordIdx, word in enumerate(bestWords):
wordPmids = wordToPmids[word]
for pmid in set(wordPmids):
pmidToWords[pmid].append(wordIdx)
svmlOutFname = join(svmlDir, db+".svml")
pmidOutFname = join(svmlDir, db+".pmids")
classOutFname = join(svmlDir, db+".classes")
logging.info("Writing to %s and %s" % (svmlOutFname, pmidOutFname))
svmlFh = open(svmlOutFname, "w")
pmidFh = open(pmidOutFname, "w")
classIdFh = open(classOutFname, "w")
allPmids = posPmids.union(negPmids)
for pmid in allPmids:
wordIdxList = pmidToWords.get(pmid, [])
if pmid in posPmids:
classId = 1
else:
classId = -1
wordVec = ["1:1"]
# svml is 1-based
# AND we reserve feature 1 to indicate if any best word matched
wordVec.extend(["%d:1" % (wordIdx+2) for wordIdx in wordIdxList])
# if no word matched, set feature 1 to 0
if len(wordVec)==1:
wordVec = ["1:0"]
wordStr = " ".join(wordVec)
svmlFh.write("%+d %s\n" % (classId, wordStr))
pmidFh.write("%d\n" % pmid)
classIdFh.write("%d\n" % classId)
svmlFh.close()
pmidFh.close()
classIdFh.close()
return svmlOutFname
def marshalToSvml(dbList, pmidListDir, dataCountFname, bestWordsDir, svmlDir):
mkEmptyDir(svmlDir)
gc.disable()
logging.info("Reading word pmids")
wordToPmids = marshal.load(open(dataCountFname))
for db in dbList:
logging.info("creating pmid -> word lists for db %s" % db)
pmidToWords = defaultdict(list)
wordFname = join(bestWordsDir, db+".txt")
pmidClasses = parsePmidClasses(pmidListDir, db)
logging.info("Getting best words from %s" % wordFname)
bestWordList = open(wordFname).read().splitlines()
#wordToIdx = dict((y,x+1) for x,y in enumerate(bestWordList))
for wordIdx, word in enumerate(bestWordList):
wordPmids = wordToPmids[word]
dbWordPmids = set(wordPmids).intersection(pmidClasses)
#wordIdx = wordToIdx[word]
for pmid in dbWordPmids:
pmidToWords[pmid].append(wordIdx)
#print pmidToWords
svmlOutFname = join(svmlDir, db+".svml")
pmidOutFname = join(svmlDir, db+".pmids")
classOutFname = join(svmlDir, db+".classes")
logging.info("Writing to %s and %s" % (svmlOutFname, pmidOutFname))
svmlFh = open(svmlOutFname, "w")
pmidFh = open(pmidOutFname, "w")
classIdFh = open(classOutFname, "w")
for pmid, wordIdxList in pmidToWords.iteritems():
classId = pmidClasses[pmid]
wordVec = ["%d:1" % (wordIdx+1) for wordIdx in wordIdxList]
wordStr = " ".join(wordVec)
svmlFh.write("%+d %s\n" % (classId, wordStr))
pmidFh.write("%d\n" % pmid)
classIdFh.write("%d\n" % classId)
gc.enable()
def splitFile(inFname, ratio, outDir1, outDir2):
" split text file and also the .pmid / .classes files "
base = basename(inFname)
baseNoExt = splitext(base)[0]
noExt = splitext(inFname)[0]
outFname1 = join(outDir1, base)
outFname2 = join(outDir2, base)
pmidInFn = noExt+".pmids"
classInFn = noExt+".classes"
logging.info("Splitting %s" % inFname)
ofhs = []
pmidFhs = []
classFhs = []
pmidFhs.append(open(join(outDir1, baseNoExt+".pmids"), "w"))
pmidFhs.append(open(join(outDir2, baseNoExt+".pmids"), "w"))
classFhs.append(open(join(outDir1, baseNoExt+".classes"), "w"))
classFhs.append(open(join(outDir2, baseNoExt+".classes"), "w"))
ofhs.append( open(outFname1, "w"))
ofhs.append( open(outFname2, "w"))
pmids = open(pmidInFn).read().splitlines()
classes = open(classInFn).read().splitlines()
for lineIdx, line in enumerate(open(inFname)):
rnd = random.random()
if rnd < ratio:
ofhIdx = 0
else:
ofhIdx = 1
ofhs[ofhIdx].write(line)
pmidFhs[ofhIdx].write(pmids[lineIdx]+"\n")
classFhs[ofhIdx].write(classes[lineIdx]+"\n")
def parseBestWords(inDir, dbList):
" parse best words and return as dict db -> list of words "
logging.info("Reading best words from %s" % inDir)
res = {}
for fname in glob.glob(join(inDir, "*.txt")):
db = splitext(basename(fname))[0]
if not db in dbList:
continue
bestWords = open(fname).read().splitlines()
res[db] = bestWords
assert(len(res)!=0)
return res
def textToSvml(textDirs, dbWords, svmlDir, addMeta, addTwoGrams, modelDir):
"""create subdirs in svmlDir, one per db and write svml files into it
use only words in bestWords and add +1/-1 classes according to dbPmids,
a dict with pmid -> list of (db, class)
"""
maxCommon.mustExistDir(svmlDir, makeDir=True)
mkEmptyDir(svmlDir)
paramDict = {
"addMeta" : addMeta,
"addTwoGrams" : addTwoGrams,
"dbWords" : dbWords,
"svmlBinDir" : svmlBinDir,
"modelDir" : modelDir
}
runner = pubGeneric.makeClusterRunner(__file__, algName="toSvml")
pubAlg.submitAnnotateWrite(runner, "wordList:SvmlWriter", textDirs, paramDict, svmlDir)
runner.finish()
logging.info("Output written to %s" % svmlDir)
def parsePmidPosNeg(db, pmidListDir):
# open list of pmids in pos and neg set as set of integers
posPmids = set([int(x) for x in open(join(pmidListDir, "pos."+db+".txt")).read().splitlines()])
negPmids = set([int(x) for x in open(join(pmidListDir, "neg."+db+".txt")).read().splitlines()])
return posPmids, negPmids
def writeWordCounts(counts, posPmids, negPmids, outFname, bestWordCount, bestWordFname):
"""
write counts of pmids in foreground / background to tab file for each word
counts is a dict word -> pmids
"""
posCount = len(posPmids)
negCount = len(negPmids)
rows = []
for word, pmidList in counts.iteritems():
pmids = set(pmidList)
posOvl = len(pmids.intersection(posPmids))
negOvl = len(pmids.intersection(negPmids))
chiSq = calcChiSq(len(pmids), posCount, negCount, posOvl, negOvl)
row = [word, len(pmids), posOvl, negOvl, chiSq]
rows.append(row)
# write best words
rows.sort(key=operator.itemgetter(-1), reverse=True)
bestWords = [row[0] for row in rows[:bestWordCount]]
ofh = open(bestWordFname, "w")
ofh.write("\n".join(bestWords))
ofh.write("\n")
ofh.close()
logging.info("Wrote %s" % bestWordFname)
ofh = open(outFname, "w")
ofh.write("# posPmids=%d, negPmids=%d\n" % (posCount, negCount))
ofh.write("\t".join(["word", "count", "posOvl", "negOvl", "chiSq"])+"\n")
for row in rows:
row = [str(s) for s in row]
ofh.write("\t".join(row)+"\n")
ofh.close()
logging.info("Wrote %s" % outFname)
return bestWords
def summarizeResults(dbList, svmlDir, categoryFname):
""" combine svml output and article ids and write categories in
a format that is easier to parse: docId<tab>dbs (comma-sep)
"""
# open outfile
ofh = open(categoryFname, "w")
ofh.write("articleId\texternalId\tpmid\tclasses\tscores\n")
logging.info("Reading directory %s" % svmlDir)
fnames = os.listdir(svmlDir)
chunkIds = set([basename(fn).split(".")[0] for fn in fnames])
logging.info("Parsing %d chunks" % len(chunkIds))
dbCounts = Counter()
pm = maxCommon.ProgressMeter(len(chunkIds))
for chunkId in chunkIds:
logging.debug("%s" % chunkId)
docClasses = defaultdict(list)
docIdRows = list(maxCommon.iterTsvRows(join(svmlDir, chunkId+".docIds")))
for db in dbList:
classFname = join(svmlDir, chunkId+".%s.classes" % db)
logging.debug("Reading class assignment from %s" % classFname)
classScores = [float(l.strip()) for l in open(classFname).readlines()]
assert(len(classScores)==len(docIdRows))
for docIdRow, classValue in itertools.izip(docIdRows, classScores):
if classValue>0.0:
docClasses[docIdRow].append((db, classValue))
dbCounts[db] += 1
logging.debug("processed %d articles" % len(docClasses))
for docIdRow, catScores in docClasses.iteritems():
artId, extId, pmid = docIdRow
catScores.sort(key=operator.itemgetter(1), reverse=True)
classes, scores = zip(*catScores) # weird python magic
scores = [str(s) for s in scores]
ofh.write("%s\t%s\t%s\t%s\t%s\n" % (artId, extId, pmid, ",".join(classes), ",".join(scores)))
pm.taskCompleted()
ofh.close()
for db, dbCount in dbCounts.iteritems():
logging.info("%s: %d assigned documents" % (db, dbCount))
logging.info("Wrote class info to %s" % categoryFname)
def makeHtmlDir(dbList, outDir):
outfname = join(outDir, "index.html")
h = html.htmlWriter(outfname)
title = "Genocoding document categories recognized by Support Vector Machine"
h.head(title)
h.h4(title)
h.startUl()
for db in dbList:
desc = pubConf.classDescriptions[db]
h.li('<a href="%s.html">%s</a> (<a href="%s.ids.txt">article IDs</a>)' % (db, desc, db))
h.endUl()
h.endHtml()
def makeHtml(dbList, pmidDir, catFname, outDir):
" "
docClasses = parseDocClasses(catFname)
mkEmptyDir(outDir)
logging.info("Writing html to %s" % outDir)
for db in dbList:
docIdFname = join(outDir, db+".ids.txt")
docFh = codecs.open(docIdFname, "w", encoding="utf8")
docFh.write("#PMID\tDOI\n")
outfname = join(outDir, db+".html")
artScores = docClasses[db]
logging.info("DB %s, filename %s" % (db, outfname))
h = html.htmlWriter(outfname)
h.head("Database: %s" % db)
desc = pubConf.classDescriptions.get(db, "")
h.h4("Documents in category <i>'%s'</i> (%s)" % (desc, db))
h.writeLn("Source database of training documents: <b>%s</b><br>" % db)
h.writeLn("Number of articles shown: %d<br>" % len(artScores))
h.writeLn('Articles already in the database are not shown<br>')
h.writeLn('<a href="%s.ids.txt">Download document IDs</a> in tab-sep format' % (db))
h.p()
h.writeLn('<hr>')
artScores.sort(key=operator.itemgetter(1), reverse=True)
pm = maxCommon.ProgressMeter(len(artScores), stepCount=100)
skippedPmids = 0
dbPmids = parsePosPmids(db, pmidDir)
for artId, score in artScores:
art = pubStore.lookupArticleByArtId(artId)
if art==None:
logging.warn("No info on artId %s?" % pmid)
continue
# don't show articles that are already in DB
if art["pmid"]!="" and int(art["pmid"]) in dbPmids:
skippedPmids += 1
continue
h.link(art["fulltextUrl"], art["title"])
h.br()
authors = art["authors"]
if len(authors)>60:
authors = authors.split(";")[0]+" et. al. "
ref = ["score: "+str(score), art["journal"], art["year"], authors]
h.writeLn(", ".join(ref))
#h.hr()
h.p()
docFh.write("%s\t%s\n" % (art["pmid"], art["doi"]))
pm.taskCompleted()
logging.info("Skipped %d PMIDs as they are already in DB" % skippedPmids)
h.endHtml()
docFh.close()
def compSvmlClasses(testSvmlDir, db):
classFname = join(testSvmlDir, db+".predClasses")
svmlFname = join(testSvmlDir, db+".classes")
print classFname, svmlFname
classes = open(classFname).read().splitlines()
svmls = open(svmlFname).read().splitlines()
TP, FN, FP, TN = 0, 0, 0, 0
realPosCount = 0
realNegCount = 0
for classLine, svmlLine in zip(classes, svmls):
predVal = float(classLine)
realVal = float(svmlLine.split()[0])
if realVal > 0:
realPosCount += 1
else:
realNegCount += 1
print realVal, predVal
if predVal<0 and realVal<0:
print "tn"
TN+=1
if predVal>0 and realVal>0:
print "tp"
TP+=1
if predVal<0 and realVal>0:
print "fp"
FP+=1