#!/usr/bin/python # -*- coding:utf-8 -*- ''' Created on 21 paź 2013 @author: mlenart ''' import os import sys import logging import codecs from morfeuszbuilder.fsa import encode from morfeuszbuilder.fsa import convertinput from morfeuszbuilder.fsa.fsa import FSA from morfeuszbuilder.fsa.serializer import Serializer, SerializationMethod from morfeuszbuilder.tagset.tagset import Tagset from morfeuszbuilder.segrules import rulesParser from morfeuszbuilder.utils import exceptions, limits from optparse import OptionParser def _checkOption(opt, parser, msg): if opt is None: print >> sys.stderr, msg parser.print_help() exit(1) def _checkCondition(cond, parser, msg): if not cond: print >> sys.stderr, msg parser.print_help() exit(1) def _parseListCallback(option, opt, value, parser): setattr(parser.values, option.dest, value.split(',')) def _checkOpen(filename, mode): try: with open(filename, mode) as _: pass if 'w' in mode: os.remove(filename) except IOError as ex: print >> sys.stderr, str(ex) exit(1) def _getDictFilename(opts, isGenerator): typeCode = 's' if isGenerator else 'a' fname = '%s-%s.dict' % (opts.dictName, typeCode) return os.path.join(opts.dictDir, fname) def _parseOptions(): """ Parses commandline args """ parser = OptionParser() parser.add_option('--input-files', type='string', dest='inputFiles', action='callback', callback=_parseListCallback, metavar='FILES', help='comma separated list of dictionary files') parser.add_option('--tagset-file', dest='tagsetFile', metavar='FILE', help='path to the file with tagset') parser.add_option('--segments-file', dest='segmentsFile', metavar='FILE', help='path to the file with segment rules') #~ parser.add_option('--trim-supneg', #~ dest='trimSupneg', #~ default=False, #~ action='store_true', #~ help='this option is ignored and exists only for backwards compatibility') parser.add_option('--dict', dest='dictName', help='the name of result dictionary') parser.add_option('--dict-dir', dest='dictDir', metavar='FILE', default=os.getcwd(), help='path to output directory (the default is current dir)') parser.add_option('--only-analyzer', dest='onlyAnalyzer', action='store_true', default=False, help='Generate dictionary for morphological analysis only (default is both analysis and synthesis)') parser.add_option('--only-generator', dest='onlyGenerator', action='store_true', default=False, help='Generate dictionary for morphological synthesis only (default is both analysis and synthesis)') parser.add_option('--analyzer-cpp', dest='analyzerCpp', metavar='FILE', help='Encode analyzer dictionary data in given c++ file') parser.add_option('--generator-cpp', dest='generatorCpp', metavar='FILE', help='Encode generator dictionary data in given c++ file') #~ parser.add_option('--use-arrays', #~ dest='useArrays', #~ action='store_true', #~ default=False, #~ help='store states reachable by 2 transitions in arrays (should speed up recognition, available only when --serialization-method=V1)') parser.add_option('--serialization-method', dest='serializationMethod', default='V1', help="FSA serialization method: \ SIMPLE - fixed-length transitions, fastest and weakest compression \ V1 - variable-length transitions, compressed labels - strongest compression \ V2 - format similar to the default in Jan Daciuk's fsa package - variable-length transitions, non-compressed labels - good compression, good speed") #~ parser.add_option('--visualize', #~ dest='visualize', #~ action='store_true', #~ default=False, #~ help='visualize result') parser.add_option('--analyzer-train-file', dest='analyzerTrainFile', help='A text file used for analyzer training. Should contain words from some large corpus - one word in each line. Resulting analysis automaton should be faster with proper train file.') parser.add_option('--generator-train-file', dest='generatorTrainFile', help='A text file used for generator training. Should contain words from some large corpus - one word in each line. Resulting synthesis automaton should be faster with proper train file.') parser.add_option('--debug', dest='debug', action='store_true', default=False, help='output some debugging info') #~ parser.add_option('--profile', #~ dest='profile', #~ action='store_true', #~ default=False, #~ help='show profiling graph (required pycallgraph and graphviz') opts, args = parser.parse_args() _checkOption(opts.inputFiles, parser, "Input file is missing") _checkOption(opts.dictDir, parser, "Output dictionary dir is missing") _checkCondition((opts.onlyAnalyzer, opts.onlyGenerator) != (True, True), parser, 'Cannot set both --only-analyzer and --only-generator') writeCpp = {opts.analyzerCpp, opts.generatorCpp} != {None} _checkCondition(opts.dictName or writeCpp, parser, "Dictionary name is missing") _checkCondition(opts.onlyGenerator or opts.analyzerCpp or not writeCpp, parser, "Analyzer .cpp output file path is missing") _checkCondition(opts.onlyAnalyzer or opts.generatorCpp or not writeCpp, parser, "Generator .cpp output file path is missing") #~ _checkCondition((opts.dictName, opts.outputCpp) != (None, None), #~ parser, 'Must set at least one of: --dict-name, --output-cpp') #~ _checkOption(opts.outputFile, parser, "Output file is missing") _checkOption(opts.tagsetFile, parser, "Tagset file is missing") _checkOption(opts.segmentsFile, parser, "Segmentation file is missing") #~ _checkOption(opts.serializationMethod, parser, "Serialization method file is missing") #~ _checkExactlyOneOptionSet([opts.analyzer, opts.generator], #~ parser, 'Must set exactly one FSA type: --analyzer or --generator') _checkOpen(opts.tagsetFile, 'r') _checkOpen(opts.segmentsFile, 'r') for filename in opts.inputFiles: _checkOpen(filename, 'r') if not opts.onlyGenerator: _checkOpen(_getDictFilename(opts, isGenerator=False), 'w') if not opts.onlyAnalyzer: _checkOpen(_getDictFilename(opts, isGenerator=True), 'w') if not opts.serializationMethod.upper() in [SerializationMethod.SIMPLE, SerializationMethod.V1]: print >> sys.stderr, '--serialization-method must be one of ('+str([SerializationMethod.SIMPLE, SerializationMethod.V1])+')' parser.print_help() exit(1) return opts def _concatFiles(inputFiles): for inputFile in inputFiles: if inputFile: with open(inputFile, 'r') as f: for line in f: if line and not ' ' in ''.join(line.split('\t')[:2]): yield line else: logging.warn(u'Ignoring line: "%s" - contains space in text form or lemma' % line.strip().decode('utf8')) def _readNamesAndQualifiers(inputFiles): names = set([u'']) qualifiers = set([frozenset()]) for line in _concatFiles(inputFiles): line = line.strip().decode('utf8') if line: _, _, _, name, qualifier = convertinput.parseLine(line) names.add(name) qualifiers.add(convertinput.parseQualifiers(qualifier)) namesMap = dict([(name, idx) for idx, name in enumerate(sorted(list(names)))]) qualifiersMap = dict([(quals, idx) for idx, quals in enumerate(sorted(qualifiers, key=lambda q: tuple(sorted(q))))]) exceptions.validate( len(qualifiersMap) <= limits.MAX_QUALIFIERS_COMBINATIONS, u'Too many qualifiers combinations. The limit is %d' % limits.MAX_QUALIFIERS_COMBINATIONS) return namesMap, qualifiersMap def _readPolimorfInput4Analyzer(inputFiles, tagset, namesMap, qualifiersMap, encoder, segmentRulesManager): logging.info('reading analyzer data from %s', str(inputFiles)) for entry in convertinput.PolimorfConverter4Analyzer(tagset, namesMap, qualifiersMap, encoder, 'utf8', segmentRulesManager).convert(_concatFiles(inputFiles)): yield entry def _readPolimorfInput4Generator(inputFiles, tagset, namesMap, qualifiersMap, encoder, segmentRulesManager): logging.info('reading generator data from %s', str(inputFiles)) for entry in convertinput.PolimorfConverter4Generator(tagset, namesMap, qualifiersMap, encoder, 'utf8', segmentRulesManager).convert(_concatFiles(inputFiles)): yield entry def _readTrainData(trainFile): with codecs.open(trainFile, 'r', 'utf8') as f: for line in f: yield line.strip() def _printStats(fsa): acceptingNum = 0 sinkNum = 0 arrayNum = 0 for s in fsa.dfs(): if s.isAccepting(): acceptingNum += 1 if s.transitionsNum == 0: sinkNum += 1 if s.serializeAsArray: arrayNum += 1 logging.info('states num: '+str(fsa.getStatesNum())) logging.info('transitions num: '+str(fsa.getTransitionsNum())) logging.info('accepting states num: '+str(acceptingNum)) logging.info('sink states num: '+str(sinkNum)) logging.info('array states num: '+str(arrayNum)) def buildAnalyzerFromPoliMorf(inputFiles, tagset, namesMap, qualifiersMap, segmentRulesManager): encoder = encode.MorphEncoder() fsa = FSA(encoder, tagset) for word, data in _readPolimorfInput4Analyzer(inputFiles, tagset, namesMap, qualifiersMap, encoder, segmentRulesManager): # print word, data fsa.addEntry(word, data) del word del data fsa.close() logging.info('------') logging.info('Analyzer FSA stats:') logging.info('------') _printStats(fsa) return fsa def buildGeneratorFromPoliMorf(inputFiles, tagset, namesMap, qualifiersMap, segmentRulesManager): encoder = encode.Encoder4Generator() fsa = FSA(encoder, tagset) inputData = _readPolimorfInput4Generator(inputFiles, tagset, namesMap, qualifiersMap, encoder, segmentRulesManager) for word, data in inputData: fsa.addEntry(word, data) fsa.close() logging.info('------') logging.info('Generator FSA stats:') logging.info('------') _printStats(fsa) return fsa def _doBuildDictionaryPart(opts, tagset, namesMap, qualifiersMap, isGenerator): logging.info('reading segmentation rules') rulesParserVersion = rulesParser.RulesParser.PARSE4ANALYZER if not isGenerator else rulesParser.RulesParser.PARSE4GENERATOR segmentRulesManager = rulesParser.RulesParser(tagset, namesMap, qualifiersMap, rulesParserVersion).parse(opts.segmentsFile) segmentationRulesData = segmentRulesManager.serialize() logging.info('done reading segmentation rules') logging.info('building automaton') buildFunction = buildAnalyzerFromPoliMorf if not isGenerator else buildGeneratorFromPoliMorf fsa = buildFunction(opts.inputFiles, tagset, namesMap, qualifiersMap, segmentRulesManager) logging.info('done building automaton') if not isGenerator and opts.analyzerTrainFile: logging.info('training analyzer automaton with '+opts.analyzerTrainFile+' ...') fsa.train(_readTrainData(opts.analyzerTrainFile)) logging.info('done training') if isGenerator and opts.generatorTrainFile: logging.info('training generator automaton with '+opts.generatorTrainFile+' ...') fsa.train(_readTrainData(opts.analyzerTrainFile)) logging.info('done training') serializer = Serializer.getSerializer(opts.serializationMethod, fsa, tagset, namesMap, qualifiersMap, segmentationRulesData) if opts.generatorCpp and isGenerator: serializer.serialize2CppFile(opts.generatorCpp, isGenerator=isGenerator) if opts.analyzerCpp and not isGenerator: serializer.serialize2CppFile(opts.analyzerCpp, isGenerator=isGenerator) if opts.dictDir: serializer.serialize2BinaryFile(_getDictFilename(opts, isGenerator=isGenerator), isGenerator=isGenerator) logging.info('total FSA size (in bytes): '+str(fsa.initialState.reverseOffset)) def main(opts): if opts.debug: logging.basicConfig(level=logging.DEBUG) else: logging.basicConfig(level=logging.INFO) logging.info('reading tagset') tagset = Tagset(opts.tagsetFile) logging.info('done reading tagset') logging.info('reading names and qualifiers') namesMap, qualifiersMap = _readNamesAndQualifiers(opts.inputFiles) logging.info('done reading names and qualifiers') if not opts.onlyGenerator: _doBuildDictionaryPart(opts, tagset, namesMap, qualifiersMap, isGenerator=False) if not opts.onlyAnalyzer: _doBuildDictionaryPart(opts, tagset, namesMap, qualifiersMap, isGenerator=True) if __name__ == '__main__': import os opts = _parseOptions() #~ try: main(opts) #~ except Exception as ex: #~ print >> sys.stderr, u'Building dictionary file failed:', unicode(ex).encode('utf8'), 'type of error:', type(ex) #~ sys.exit(1) #~ finally: #~ pass