def handleOneIndex(indexpath, subdir, indexname, fast):
print(indexpath, subdir, indexname)
indexstatuspath = indexpath + config.getStatusPostfix()
indexstatus = utils.load_status(indexstatuspath)
if not utils.check_epoch(indexstatus, 'Prepare'):
raise utils.EpochError('Please prepare first.\n')
if utils.check_epoch(indexstatus, 'Populate'):
return
workdir = config.getWordRecognizerDir() + os.sep + \
subdir + os.sep + indexname
print(workdir)
shmdir = config.getInMemoryFileSystem()
for i in range(1, N + 1):
if fast:
#copy file
filename = config.getNgramFileName(i)
filepath = workdir + os.sep + filename
shmfilepath = shmdir + os.sep + filename
utils.copyfile(filepath, shmfilepath)
handleOnePass(indexpath, shmdir, i)
pruneNgramTable(indexpath, shmdir, i)
utils.copyfile(shmfilepath, filepath)
os.unlink(shmfilepath)
else:
handleOnePass(indexpath, workdir, i)
pruneNgramTable(indexpath, workdir, i)
#sign epoch
utils.sign_epoch(indexstatus, 'Populate')
utils.store_status(indexstatuspath, indexstatus)
def generateOneText(infile, modelfile, reportfile):
infilestatuspath = infile + config.getStatusPostfix()
infilestatus = utils.load_status(infilestatuspath)
if not utils.check_epoch(infilestatus, 'MergeSequence'):
raise utils.EpochError('Please mergeseq first.\n')
if utils.check_epoch(infilestatus, 'Generate'):
return False
#begin processing
cmdline = ['../utils/training/gen_k_mixture_model', \
'--maximum-occurs-allowed', \
str(config.getMaximumOccursAllowed()), \
'--maximum-increase-rates-allowed', \
str(config.getMaximumIncreaseRatesAllowed()), \
'--k-mixture-model-file', \
modelfile, infile + \
config.getMergedPostfix()]
subprocess = Popen(cmdline, shell=False, stderr=PIPE, \
close_fds=True)
lines = subprocess.stderr.readlines()
if lines:
print('found error report')
with open(reportfile, 'ab') as f:
f.writelines(lines)
(pid, status) = os.waitpid(subprocess.pid, 0)
if status != 0:
sys.exit('gen_k_mixture_model encounters error.')
#end processing
utils.sign_epoch(infilestatus, 'Generate')
utils.store_status(infilestatuspath, infilestatus)
return True
def handleOneIndex(indexpath):
indexstatuspath = indexpath + config.getStatusPostfix()
indexstatus = utils.load_status(indexstatuspath)
if not utils.check_epoch(indexstatus, 'Segment'):
raise utils.EpochError('Please segment first.\n')
if utils.check_epoch(indexstatus, 'MergeSequence'):
return
#begin processing
indexfile = open(indexpath, 'r')
for oneline in indexfile.readlines():
#remove tailing '\n'
oneline = oneline.rstrip(os.linesep)
(title, textpath) = oneline.split('#')
infile = config.getTextDir() + textpath
outfile = config.getTextDir() + textpath + config.getMergedPostfix()
reportfile = config.getTextDir() + textpath + \
config.getMergedReportPostfix()
print("Processing " + title + '#' + textpath)
mergeOneText(infile, outfile, reportfile)
print("Processed " + title + '#' + textpath)
indexfile.close()
#end processing
utils.sign_epoch(indexstatus, 'MergeSequence')
utils.store_status(indexstatuspath, indexstatus)
def mergeOneText(infile, outfile, reportfile):
infilestatuspath = infile + config.getStatusPostfix()
infilestatus = utils.load_status(infilestatuspath)
if not utils.check_epoch(infilestatus, 'Segment'):
raise utils.EpochError('Please segment first.\n')
if utils.check_epoch(infilestatus, 'MergeSequence'):
return
infile = infile + config.getSegmentPostfix()
#begin processing
cmdline = ['../utils/segment/mergeseq', \
'-o', outfile, infile]
subprocess = Popen(cmdline, shell=False, stderr=PIPE, \
close_fds=True)
lines = subprocess.stderr.readlines()
if lines:
print('found error report')
with open(reportfile, 'wb') as f:
f.writelines(lines)
os.waitpid(subprocess.pid, 0)
#end processing
utils.sign_epoch(infilestatus, 'MergeSequence')
utils.store_status(infilestatuspath, infilestatus)
def handleOneDocument(infile, cur, length):
print(infile, length)
infilestatuspath = infile + config.getStatusPostfix()
infilestatus = utils.load_status(infilestatuspath)
if not utils.check_epoch(infilestatus, 'Segment'):
raise utils.EpochError('Please segment first.\n')
if utils.check_epoch(infilestatus, 'Populate'):
return False
sep = config.getWordSep()
#train
docfile = open(infile + config.getSegmentPostfix(), 'r')
words = []
for oneline in docfile.readlines():
oneline = oneline.rstrip(os.linesep)
if len(oneline) == 0:
continue
(token, word) = oneline.split(" ", 1)
token = int(token)
if 0 == token:
words = []
else:
words.append(word)
if len(words) < length:
continue
if len(words) > length:
words.pop(0)
assert len(words) == length
#do sqlite training
words_str = sep + sep.join(words) + sep
#print(words_str)
rowcount = cur.execute(UPDATE_NGRAM_DML, (words_str, )).rowcount
#print(rowcount)
assert rowcount <= 1
if 0 == rowcount:
cur.execute(INSERT_NGRAM_DML, (words_str, ))
docfile.close()
#sign epoch only after last pass
if N == length:
utils.sign_epoch(infilestatus, 'Populate')
utils.store_status(infilestatuspath, infilestatus)
return True
def handleOneIndex(indexpath, subdir, indexname):
print(indexpath, subdir, indexname)
indexstatuspath = indexpath + config.getStatusPostfix()
indexstatus = utils.load_status(indexstatuspath)
if not utils.check_epoch(indexstatus, 'NewWord'):
raise utils.EpochError('Please new word first.\n')
if utils.check_epoch(indexstatus, 'MarkPinyin'):
return
workdir = config.getWordRecognizerDir() + os.sep + \
subdir + os.sep + indexname
print(workdir)
markPinyins(workdir)
#sign epoch
utils.sign_epoch(indexstatus, 'MarkPinyin')
utils.store_status(indexstatuspath, indexstatus)
def handleOneModel(modelfile, reportfile):
modelfilestatuspath = modelfile + config.getStatusPostfix()
modelfilestatus = utils.load_status(modelfilestatuspath)
if not utils.check_epoch(modelfilestatus, 'Generate'):
raise utils.EpochError('Please generate first.\n')
if utils.check_epoch(modelfilestatus, 'Estimate'):
return
reporthandle = open(reportfile, 'wb')
result_line_prefix = "average lambda:"
avg_lambda = 0.
#begin processing
cmdline = ['../utils/training/estimate_k_mixture_model', \
'--deleted-bigram-file', \
config.getEstimatesModel(), \
'--bigram-file', \
modelfile]
subprocess = Popen(cmdline, shell=False, stdout=PIPE, \
close_fds=True)
for line in subprocess.stdout.readlines():
reporthandle.writelines([line])
#remove trailing '\n'
line = line.decode('utf-8')
line = line.rstrip(os.linesep)
if line.startswith(result_line_prefix):
avg_lambda = float(line[len(result_line_prefix):])
reporthandle.close()
(pid, status) = os.waitpid(subprocess.pid, 0)
if status != 0:
sys.exit('estimate k mixture model returns error.')
#end processing
print('average lambda:', avg_lambda)
modelfilestatus['EstimateScore'] = avg_lambda
utils.sign_epoch(modelfilestatus, 'Estimate')
utils.store_status(modelfilestatuspath, modelfilestatus)
def handleOneIndex(indexpath, subdir, indexname):
print(indexpath, subdir, indexname)
indexstatuspath = indexpath + config.getStatusPostfix()
indexstatus = utils.load_status(indexstatuspath)
if not utils.check_epoch(indexstatus, 'Segment'):
raise utils.EpochError('Please segment first.\n')
if utils.check_epoch(indexstatus, 'Prepare'):
return
#create directory
onedir = config.getWordRecognizerDir() + os.sep + \
subdir + os.sep + indexname
os.path.exists(onedir) or os.makedirs(onedir)
#create sqlite databases
createSqliteDatabases(onedir)
#sign epoch
utils.sign_epoch(indexstatus, 'Prepare')
utils.store_status(indexstatuspath, indexstatus)
def handleOneIndex(indexpath, subdir, indexname):
print(indexpath, subdir, indexname)
indexstatuspath = indexpath + config.getStatusPostfix()
indexstatus = utils.load_status(indexstatuspath)
if not utils.check_epoch(indexstatus, 'PartialWord'):
raise utils.EpochError('Please partial word first.\n')
if utils.check_epoch(indexstatus, 'NewWord'):
return
workdir = config.getWordRecognizerDir() + os.sep + \
subdir + os.sep + indexname
print(workdir)
createBigramSqlite(workdir)
populateBigramSqlite(workdir)
filename = config.getBigramFileName()
filepath = workdir + os.sep + filename
conn = sqlite3.connect(filepath)
prethres = computeThreshold(conn, "prefix")
indexstatus['NewWordPrefixThreshold'] = prethres
postthres = computeThreshold(conn, "postfix")
indexstatus['NewWordPostfixThreshold'] = postthres
utils.store_status(indexstatuspath, indexstatus)
filterPartialWord(workdir, conn, prethres, postthres)
conn.commit()
if conn:
conn.close()
#sign epoch
utils.sign_epoch(indexstatus, 'NewWord')
utils.store_status(indexstatuspath, indexstatus)
def handleOneIndex(indexpath, subdir, indexname):
print(indexpath, subdir, indexname)
indexstatuspath = indexpath + config.getStatusPostfix()
indexstatus = utils.load_status(indexstatuspath)
if not utils.check_epoch(indexstatus, 'Populate'):
raise utils.EpochError('Please populate first.\n')
if utils.check_epoch(indexstatus, 'PartialWord'):
return
workdir = config.getWordRecognizerDir() + os.sep + \
subdir + os.sep + indexname
print(workdir)
threshold = getThreshold(workdir)
indexstatus['PartialWordThreshold'] = threshold
utils.store_status(indexstatuspath, indexstatus)
recognizePartialWord(workdir, threshold)
#sign epoch
utils.sign_epoch(indexstatus, 'PartialWord')
utils.store_status(indexstatuspath, indexstatus)
def gatherModels(path, indexname):
indexfilestatuspath = indexname + config.getStatusPostfix()
indexfilestatus = utils.load_status(indexfilestatuspath)
if utils.check_epoch(indexfilestatuspath, 'Estimate'):
return
#begin processing
indexfile = open(indexname, "w")
for root, dirs, files in os.walk(path, topdown=True, onerror=handleError):
for onefile in files:
filepath = os.path.join(root, onefile)
if onefile.endswith(config.getModelPostfix()):
#append one record to index file
subdir = os.path.relpath(root, path)
statusfilepath = filepath + config.getStatusPostfix()
status = utils.load_status(statusfilepath)
if not (utils.check_epoch(status, 'Estimate') and \
'EstimateScore' in status):
raise utils.EpochError('Unknown Error:\n' + \
'Try re-run estimate.\n')
avg_lambda = status['EstimateScore']
line = subdir + '#' + onefile + '#' + str(avg_lambda)
indexfile.writelines([line, os.linesep])
#record written
elif onefile.endswith(config.getStatusPostfix()):
pass
elif onefile.endswith(config.getIndexPostfix()):
pass
elif onefile.endswith(config.getReportPostfix()):
pass
else:
print('Unexpected file:' + filepath)
indexfile.close()
#end processing
utils.sign_epoch(indexfilestatus, 'Estimate')
utils.store_status(indexfilestatuspath, indexfilestatus)
default=config.getFinalModelDir())
parser.add_argument('tryname', action='store', \
help='the storage directory')
args = parser.parse_args()
print(args)
tryname = 'try' + args.tryname
trydir = os.path.join(args.finaldir, tryname)
if not os.access(trydir, os.F_OK):
sys.exit(tryname + "doesn't exist.")
cwdstatuspath = os.path.join(trydir, config.getFinalStatusFileName())
cwdstatus = utils.load_status(cwdstatuspath)
if not utils.check_epoch(cwdstatus, 'Prune'):
raise utils.EpochError('Please tryprune first.')
if utils.check_epoch(cwdstatus, 'Evaluate'):
sys.exit('already evaluated.')
print('checking')
checkData()
modelfile = os.path.join(trydir, config.getFinalModelFileName())
destfile = os.path.join(libpinyin_dir, 'data', \
config.getFinalModelFileName())
utils.copyfile(modelfile, destfile)
print('cleaning')
cleanUpData()
def handleOneIndex(indexpath, subdir, indexname, fast):
inMemoryFile = "model.db"
modeldir = os.path.join(config.getModelDir(), subdir, indexname)
os.path.exists(modeldir) or os.makedirs(modeldir)
def cleanupInMemoryFile():
modelfile = os.path.join(config.getInMemoryFileSystem(), inMemoryFile)
reportfile = modelfile + config.getReportPostfix()
if os.access(modelfile, os.F_OK):
os.unlink(modelfile)
if os.access(reportfile, os.F_OK):
os.unlink(reportfile)
def copyoutInMemoryFile(modelfile):
inmemoryfile = os.path.join\
(config.getInMemoryFileSystem(), inMemoryFile)
inmemoryreportfile = inmemoryfile + config.getReportPostfix()
reportfile = modelfile + config.getReportPostfix()
if os.access(inmemoryfile, os.F_OK):
utils.copyfile(inmemoryfile, modelfile)
if os.access(inmemoryreportfile, os.F_OK):
utils.copyfile(inmemoryreportfile, reportfile)
def cleanupFiles(modelnum):
modeldir = os.path.join(config.getModelDir(), subdir, indexname)
modelfile = os.path.join( \
modeldir, config.getCandidateModelName(modelnum))
reportfile = modelfile + config.getReportPostfix()
if os.access(modelfile, os.F_OK):
os.unlink(modelfile)
if os.access(reportfile, os.F_OK):
os.unlink(reportfile)
def storeModelStatus(modelfile, textnum, nexttextnum):
#store model info in status file
modelstatuspath = modelfile + config.getStatusPostfix()
#create None status
modelstatus = {}
modelstatus['GenerateStart'] = textnum
modelstatus['GenerateEnd'] = nexttextnum
utils.sign_epoch(modelstatus, 'Generate')
utils.store_status(modelstatuspath, modelstatus)
print(indexpath, subdir, indexname)
indexstatuspath = indexpath + config.getStatusPostfix()
indexstatus = utils.load_status(indexstatuspath)
if not utils.check_epoch(indexstatus, 'MergeSequence'):
raise utils.EpochError('Please mergeseq first.\n')
if utils.check_epoch(indexstatus, 'Generate'):
return
#continue generating
textnum, modelnum, aggmodelsize = 0, 0, 0
if 'GenerateTextEnd' in indexstatus:
textnum = indexstatus['GenerateTextEnd']
if 'GenerateModelEnd' in indexstatus:
modelnum = indexstatus['GenerateModelEnd']
#clean up previous file
if fast:
cleanupInMemoryFile()
cleanupFiles(modelnum)
#begin processing
indexfile = open(indexpath, 'r')
for i, oneline in enumerate(indexfile.readlines()):
#continue last generating
if i < textnum:
continue
#remove trailing '\n'
oneline = oneline.rstrip(os.linesep)
(title, textpath) = oneline.split('#')
infile = config.getTextDir() + textpath
infilesize = utils.get_file_length(infile + config.getMergedPostfix())
if infilesize < config.getMinimumFileSize():
print("Skipping " + title + '#' + textpath)
continue
if fast:
modelfile = os.path.join(config.getInMemoryFileSystem(), \
inMemoryFile)
else:
modelfile = os.path.join(modeldir, \
config.getCandidateModelName(modelnum))
reportfile = modelfile + config.getReportPostfix()
print("Proccessing " + title + '#' + textpath)
if generateOneText(infile, modelfile, reportfile):
aggmodelsize += infilesize
print("Processed " + title + '#' + textpath)
if aggmodelsize > config.getCandidateModelSize():
#copy out in memory file
if fast:
modelfile = os.path.join\
(modeldir, config.getCandidateModelName(modelnum))
copyoutInMemoryFile(modelfile)
cleanupInMemoryFile()
#the model file is in disk now
nexttextnum = i + 1
storeModelStatus(modelfile, textnum, nexttextnum)
#new model candidate
aggmodelsize = 0
textnum = nexttextnum
modelnum += 1
#clean up next file
cleanupFiles(modelnum)
#save current progress in status file
indexstatus['GenerateTextEnd'] = nexttextnum
indexstatus['GenerateModelEnd'] = modelnum
utils.store_status(indexstatuspath, indexstatus)
#copy out in memory file
if fast:
modelfile = os.path.join\
(modeldir, config.getCandidateModelName(modelnum))
copyoutInMemoryFile(modelfile)
cleanupInMemoryFile()
#the model file is in disk now
nexttextnum = i + 1
storeModelStatus(modelfile, textnum, nexttextnum)
indexfile.close()
#end processing
#save current progress in status file
modelnum += 1
indexstatus['GenerateTextEnd'] = nexttextnum
indexstatus['GenerateModelEnd'] = modelnum
utils.sign_epoch(indexstatus, 'Generate')
utils.store_status(indexstatuspath, indexstatus)