def loadDicts(prefix = 'cmj'):
global docids, rdocids, tokenids
docids = [item.strip().split('\t') for item in open(common.dataFile(prefix + '_docids.txt')).readlines()]
docids = dict([(int(id), val) for id, val in docids]) # map int -> int_id
rdocids = dict([(v, k) for k, v in docids.iteritems()]) # map int_id -> int
tokenids = [item.strip().split('\t') for item in codecs.open(common.dataFile(prefix + '_tokenids.txt'), 'r', 'utf8').readlines()]
tokenids = dict([(int(id), val) for id, val in tokenids])
def create_refsdb(baseDir, downloadMSC = False):
"""read references from baseDir files and merge them together into an ArticleDB database.
also check for consistency and print some stats
store the resulting database on disk as 'refs_baseDir.pdl'.
"""
import cPickle
reffiles = utils_iddb.getRefFiles(common.inputPath(baseDir))
cPickle.dump(reffiles, open(common.dataFile('reffiles.pkl'), 'w'), protocol = -1) # store filesystem paths to references.xml files
refdb = {}
for reffile, refid in reffiles.iteritems():
refs = utils_iddb.parseReferences(reffile, downloadMSC = downloadMSC)
refdb[refid] = refs
logging.info('id=%s: retrieved %i references' % (refid, len(refs)))
if downloadMSC:
cPickle.dump(refdb, open(common.dataFile('refs_partial.pkl'), 'w'), protocol = -1) # dump database immediately after each iteration
f = open(common.dataFile('refs.pkl'), 'w')
cPickle.dump(refdb, f, protocol = -1) # dump database immediately after each iteration
f.close()
#print some statistics
logging.info("%i MSC download attemps (%i ok, %i failed)" % (utils_iddb.attempts_all, utils_iddb.attempts_success, utils_iddb.attempts_failed))
logging.info('reference database size: %i references in %i articles' % (sum([len(refs) for refs in refdb.itervalues()]), len(refdb)))
db = ArticleDB.ArticleDB(common.dbFile('refs', baseDir), mode = 'override')
insert_errors = 0
for id, reflist in refdb.iteritems():
for num, ref in enumerate(reflist):
ref.id_int = id + ':' + str(num + 1) # references.xml reference counting starts with '1'
if not db.insertArticle(ref):
insert_errors += 1
#print '.',
db.commit()
logging.info('resulting database has %i records (originally %i)' % (len(db), sum([len(refs) for refs in refdb.itervalues()])))
logging.info('detected %i inconsistency collisions' % insert_errors)
def makeDefs():
deffile = common.dataFile('mscdefs.txt')
result = {}
for mscdef in open(deffile):
result[mscdef[:2]] = mscdef
import cPickle
f = open(common.dataFile('mscdefs.pkl'), 'w')
cPickle.dump(result, f)
f.close()
return result
def loadDicts(prefix='cmj'):
global docids, rdocids, tokenids
docids = [
item.strip().split('\t')
for item in open(common.dataFile(prefix + '_docids.txt')).readlines()
]
docids = dict([(int(id), val) for id, val in docids]) # map int -> int_id
rdocids = dict([(v, k)
for k, v in docids.iteritems()]) # map int_id -> int
tokenids = [
item.strip().split('\t')
for item in codecs.open(common.dataFile(prefix + '_tokenids.txt'), 'r',
'utf8').readlines()
]
tokenids = dict([(int(id), val) for id, val in tokenids])
def getArts(fname, minFreq):
db = ArticleDB.ArticleDB(common.dataFile(fname), mode = 'open')
import ipyutils
language = 'eng'
ipyutils.loadDicts(prefix = 'gensim_' + language)
arts = [Article.Article(rec) for rec in db.db if rec['language'] == language]
for art in arts:
art.msc = list(set(mscs.niceMSC(msc, prefix = 2)[0] for msc in art.msc))
logging.info('loaded %i articles from %s' % (len(arts), fname))
arts = [art for art in arts if len(art.msc)==1 and art.body and art.id_int in ipyutils.rdocids]
logging.info('extracted %i articles with exactly one MSC and non-empty body' % (len(arts)))
okmsc = mscs.getFreqMSCs(arts, minFreq, useMains = True)
mscs.printStats(arts)
for art in arts:
art.body = art.body.decode('utf8')
# art.body = art.body.decode('ascii', 'ignore')
art.body = art.body.translate(BAD_MAP).encode('utf8')
if art.title:
art.title = art.title.decode('utf8')
# art.title = art.title.decode('ascii', 'ignore')
art.title = art.title.translate(BAD_MAP).encode('utf8')
art.msc = [mscs.niceMSC(art.msc[0], prefix = 2)[0]]
arts = [art for art in arts if art.msc[0] in okmsc]
allmsc, mainmsc = mscs.getMSCCnts(arts)
for msc, mscarts in allmsc.iteritems():
logging.info("class %s: %i articles" % (msc, len(mscarts)))
logging.info("======================")
logging.info("sum: %i articles" % sum(len(mscarts) for mscarts in allmsc.itervalues()))
logging.debug('using %i articles from all %s msc classes that are covered by at least %i articles' % (len(arts), sorted(list(okmsc)), minFreq))
return arts
def loadArts(dbFile = 'main_cmj.pdl'):
import ArticleDB
import Article
import common
global db, arts
db = ArticleDB.ArticleDB(common.dataFile(dbFile), mode = 'open')
arts = [Article.Article(rec) for rec in db.db if rec['id_int'] in rdocids]
def saveAsCorpus(arts, fname, fnc = lambda c: c):
logging.info('saving corpus as %s' % fname)
f = open(common.dataFile(fname), 'w')
f.write('<?xml version="1.0" encoding="utf-8" ?>\n')
f.write('<articles>\n')
for art in arts:
f.write('<article id="%s" lang="%s">\n' % (art.id_int, art.language))
if art.msc:
f.write('<category>\n')
# assert len(art.msc) == 1
f.write(art.msc[0])
f.write('\n</category>\n')
if art.title:
f.write('<title>\n')
f.write(fnc(art.title))
f.write('\n</title>\n')
if art.body:
f.write('<text>\n')
f.write(fnc(art.body))
f.write('\n</text>\n')
if art.references:
f.write('<references>\n')
f.write(art.body)
f.write('\n</references>\n')
f.write('</article>\n')
f.write('</articles>\n')
f.close()
def loadArts(dbFile='main_cmj.pdl'):
import ArticleDB
import Article
import common
global db, arts
db = ArticleDB.ArticleDB(common.dataFile(dbFile), mode='open')
arts = [Article.Article(rec) for rec in db.db if rec['id_int'] in rdocids]
def saveAsCorpus(arts, fname, fnc=lambda c: c):
logging.info('saving corpus as %s' % fname)
f = open(common.dataFile(fname), 'w')
f.write('<?xml version="1.0" encoding="utf-8" ?>\n')
f.write('<articles>\n')
for art in arts:
f.write('<article id="%s" lang="%s">\n' % (art.id_int, art.language))
if art.msc:
f.write('<category>\n')
# assert len(art.msc) == 1
f.write(art.msc[0])
f.write('\n</category>\n')
if art.title:
f.write('<title>\n')
f.write(fnc(art.title))
f.write('\n</title>\n')
if art.body:
f.write('<text>\n')
f.write(fnc(art.body))
f.write('\n</text>\n')
if art.references:
f.write('<references>\n')
f.write(art.body)
f.write('\n</references>\n')
f.write('</article>\n')
f.write('</articles>\n')
f.close()
def loadMsc2Id(lang):
cats = [
item.strip().split('\t')
for item in open(common.dataFile('serial_mscids_%s.txt' %
lang)).readlines()
]
cats = dict([(int(id), val) for id, val in cats]) # map int -> id
rcats = dict([(v, k) for k, v in cats.iteritems()]) # map id -> int
return cats, rcats
def create_refsdb(baseDir, downloadMSC=False):
"""read references from baseDir files and merge them together into an ArticleDB database.
also check for consistency and print some stats
store the resulting database on disk as 'refs_baseDir.pdl'.
"""
import cPickle
reffiles = utils_iddb.getRefFiles(common.inputPath(baseDir))
cPickle.dump(reffiles,
open(common.dataFile('reffiles.pkl'), 'w'),
protocol=-1) # store filesystem paths to references.xml files
refdb = {}
for reffile, refid in reffiles.iteritems():
refs = utils_iddb.parseReferences(reffile, downloadMSC=downloadMSC)
refdb[refid] = refs
logging.info('id=%s: retrieved %i references' % (refid, len(refs)))
if downloadMSC:
cPickle.dump(
refdb,
open(common.dataFile('refs_partial.pkl'), 'w'),
protocol=-1) # dump database immediately after each iteration
f = open(common.dataFile('refs.pkl'), 'w')
cPickle.dump(refdb, f,
protocol=-1) # dump database immediately after each iteration
f.close()
#print some statistics
logging.info("%i MSC download attemps (%i ok, %i failed)" %
(utils_iddb.attempts_all, utils_iddb.attempts_success,
utils_iddb.attempts_failed))
logging.info('reference database size: %i references in %i articles' %
(sum([len(refs) for refs in refdb.itervalues()]), len(refdb)))
db = ArticleDB.ArticleDB(common.dbFile('refs', baseDir), mode='override')
insert_errors = 0
for id, reflist in refdb.iteritems():
for num, ref in enumerate(reflist):
ref.id_int = id + ':' + str(
num + 1) # references.xml reference counting starts with '1'
if not db.insertArticle(ref):
insert_errors += 1
#print '.',
db.commit()
logging.info('resulting database has %i records (originally %i)' %
(len(db), sum([len(refs) for refs in refdb.itervalues()])))
logging.info('detected %i inconsistency collisions' % insert_errors)
def printStats(arts):
"""print some MSC stats for an article database (ArticleDB)"""
mscs, mains = getMSCCnts(arts)
uniq_mscs = set(mscs.keys())
logging.info("#categories present in the db = %i" % len(uniq_mscs))
# print uniq_mscs
mscdefs = set(makeDefs().keys())
if not uniq_mscs.issubset(mscdefs):
# print 'db:', uniq_mscs
# print 'defs:', mscdefs
logging.warning(
"unrecognized MSC 2-digit code(s) present in the database: %s" %
sorted(uniq_mscs - mscdefs))
# uniq_mscs = uniq_mscs.union(mscdefs)
logging.info('id\ttotal\tprimary')
logging.info('==============================')
for msc in sorted(list(uniq_mscs)):
logging.info("%s\t%i\t%i" % (msc, len(mscs.setdefault(
msc, [])), len(mains.setdefault(msc, []))))
logging.info('==============================')
len_mscs = [len(val) for val in mscs.itervalues()]
len_mains = [len(val) for val in mains.itervalues()]
logging.info('avg\t%i\t%i' %
(sum(len_mscs) / len(mscs), sum(len_mains) / len(mscs)))
logging.info(
'median\t%i\t%i' %
(sorted(len_mscs)[len(mscs) / 2], sorted(len_mains)[len(mains) / 2]))
lens = [len(art.msc) for art in arts if art.msc != None]
logging.info('average MSC codes per article = %.2f' %
(1.0 * sum(lens) / len(arts)))
import cPickle
cPickle.dump(mscs, open(common.dataFile('mscs_all.pkl'), 'w'), protocol=-1)
cPickle.dump(mains,
open(common.dataFile('mscs_primary.pkl'), 'w'),
protocol=-1)
def getArts(fname, minFreq):
db = ArticleDB.ArticleDB(common.dataFile(fname), mode='open')
import ipyutils
language = 'eng'
ipyutils.loadDicts(prefix='gensim_' + language)
arts = [
Article.Article(rec) for rec in db.db if rec['language'] == language
]
for art in arts:
art.msc = list(set(mscs.niceMSC(msc, prefix=2)[0] for msc in art.msc))
logging.info('loaded %i articles from %s' % (len(arts), fname))
arts = [
art for art in arts
if len(art.msc) == 1 and art.body and art.id_int in ipyutils.rdocids
]
logging.info(
'extracted %i articles with exactly one MSC and non-empty body' %
(len(arts)))
okmsc = mscs.getFreqMSCs(arts, minFreq, useMains=True)
mscs.printStats(arts)
for art in arts:
art.body = art.body.decode('utf8')
# art.body = art.body.decode('ascii', 'ignore')
art.body = art.body.translate(BAD_MAP).encode('utf8')
if art.title:
art.title = art.title.decode('utf8')
# art.title = art.title.decode('ascii', 'ignore')
art.title = art.title.translate(BAD_MAP).encode('utf8')
art.msc = [mscs.niceMSC(art.msc[0], prefix=2)[0]]
arts = [art for art in arts if art.msc[0] in okmsc]
allmsc, mainmsc = mscs.getMSCCnts(arts)
for msc, mscarts in allmsc.iteritems():
logging.info("class %s: %i articles" % (msc, len(mscarts)))
logging.info("======================")
logging.info("sum: %i articles" %
sum(len(mscarts) for mscarts in allmsc.itervalues()))
logging.debug(
'using %i articles from all %s msc classes that are covered by at least %i articles'
% (len(arts), sorted(list(okmsc)), minFreq))
return arts
def saveMsc2Id(cats, lang):
DocumentCollection.saveLex(cats,
common.dataFile('serial_mscids_%s.txt' % lang))
def loadMSCdict(fname):
import cPickle
mscs = cPickle.load(open(common.dataFile(fname), 'r'))
return mscs
old.append(art)
mscs[top] = old
if isMain:
old = mains.get(top, [])
old.append(art)
mains[top] = old
return mscs, mains
def getFreqMSCs(arts, minFreq=50, useMains=False):
mscs, mains = getMSCCnts(arts)
if useMains:
mscs = mains
uniqMSCs = set(mscs.keys())
result = set(
[msc for msc in sorted(list(uniqMSCs)) if len(mscs[msc]) >= minFreq])
# print [(msc, len(mscs[msc])) for msc in result]
return result
if __name__ == "__main__":
logging.basicConfig(level=logging.DEBUG)
import ArticleDB
# print sorted(makeDefs().keys())
db = ArticleDB.ArticleDB(common.dataFile('tex_casopis.pdl'), mode='open')
arts = [Article.Article(rec) for rec in db.db if rec['msc']]
# print len(db)
logging.info('gathering MSC stats from %i articles' % len(arts))
printStats(arts)
def loadMSCdict(fname):
import cPickle
mscs = cPickle.load(open(common.dataFile(fname), 'r'))
return mscs
def saveMsc2Id(cats, lang):
DocumentCollection.saveLex(cats, common.dataFile("serial_mscids_%s.txt" % lang))
def loadMsc2Id(lang):
cats = [item.strip().split("\t") for item in open(common.dataFile("serial_mscids_%s.txt" % lang)).readlines()]
cats = dict([(int(id), val) for id, val in cats]) # map int -> id
rcats = dict([(v, k) for k, v in cats.iteritems()]) # map id -> int
return cats, rcats
