def extractDialogWithSudachi(rootDir):
outputDir = utils.getOutputPath(rootDir, "stats")
tokenizer_obj = dictionary.Dictionary().create()
mode = tokenizer.Tokenizer.SplitMode.C
unigrams = []
bigrams = []
trigrams = []
fourgrams = []
POS_LIST = ["名詞", "動詞", "副詞", "形容詞", "連体詞", "形状詞"]
for fn, fd in utils.loadFiles(rootDir):
for line in fd:
line = line.strip()
wordList = []
for word in tokenizer_obj.tokenize(line, mode):
if word.part_of_speech()[0] not in POS_LIST:
continue
wordList.append(
(word.dictionary_form(), word.part_of_speech()[0]))
print([
word.surface(),
word.dictionary_form(),
word.part_of_speech()[0]
])
unigrams.extend(getChunks(wordList, 1))
bigrams.extend(getChunks(wordList, 2))
trigrams.extend(getChunks(wordList, 3))
fourgrams.extend(getChunks(wordList, 4))
_output(outputDir, unigrams, bigrams, trigrams, fourgrams)
def collect_tg_execution(self, _dirpath, _outputname, _args):
# parameter passing
_pattern = _args.pattern if _args.pattern is not None else r'(\d+)-TG_ESAIL_SAFE_T(\d)_parallel.log'
# _pattern = _args.pattern if _args.pattern is not None else r'(\d+)-TG_ESAIL_SAFE_bf(\d)_parallel.log'
_nParition = _args.Nums if _args.Nums is not None else 5
# listing target directories
files = utils.loadFiles([{'path':_dirpath}], _ptn=_pattern, _sort=True)
# output
output = open(_outputname, "w")
output.write("JobID,PartitionID,seq,time\n")
# progressing
progress = tqdm(desc='Collecting data', total=len(files), unit=' #', postfix=None)
prev = 0
cnt = 0
for item in files:
if prev==int(item[1]):
cnt += 1
else:
cnt = 0
prev = int(item[1])
try:
header = '%d, %s' % (item['jobID'], item[1])
self.collect_logtime(item['path'], output, header, cnt*10)
except Exception as e:
print('Filed to get information: run%02d' % int(item['Run']))
progress.update(1)
progress.set_postfix_str(item['path'])
progress.close()
output.close()
def extractDialog(rootDir):
outputDir = utils.getOutputPath(rootDir, "dialog")
for fn, fd in utils.loadFiles(rootDir):
sentenceList = []
for line in fd:
if ":「" not in line:
continue
line = line.split(":「")[1].rstrip().rstrip("」")
for punctuation in NON_FINAL_PUNCTUATION:
line = line.replace(punctuation, "")
tmpSentenceList = [line]
for punctuation in SENTENCE_FINAL_PUNCTUATION:
subTmpSentenceList = []
for sentence in tmpSentenceList:
subTmpSentenceList.extend(sentence.split(punctuation))
tmpSentenceList = [
line.strip() for line in subTmpSentenceList if line != ""
]
sentenceList.extend(tmpSentenceList)
with io.open(os.path.join(outputDir, fn), "w", encoding="utf-8") as fd:
for line in sentenceList:
fd.write(line + "\n")
def clean(root, outputPath):
"""
A little bit involved but with a bit of planning, it wasn't too difficult.
Fortunately, three games all used the same formatting, so this function
yielded a lot of text.
"""
END_OF_HEADER = "TIMESTAMPS"
nameRe = re.compile("\n{2,}.*?\n[（「１]")
if not os.path.exists(outputPath):
os.mkdir(outputPath)
for fn, fd in utils.loadFiles(root):
data = fd.read()
start = data.find(END_OF_HEADER) + len(END_OF_HEADER) + 1
data = data[start:].lstrip()
# Remove lines containing an arrow in them
data = removeLinesContaining(data, "→")
start = nameRe.search(data).start()
preamble = getPreamble(data[:start])
data = data[start:]
data = removeFooter(data)
# Protect events by making them look like speech
data = data.replace("\n・", "イベント\n「・")
data = data.replace("\n\n１．", "\n\n選択\n「１．")
epilogue = getEpilogue(data)
outputList = [
preamble,
] + splitDataBySpeechEvent(data)
outputList.append(epilogue)
outputData = "\n".join(outputList) + "\n"
outputData = outputData.replace(" ", "") # Remove ideographic space
outputData = setSpeakerBeforeDecisions(outputData)
# Clean up the earlier protected 'events'
outputData = removePlaceholderSpeech(outputData, "イベント:「・", "\n:・")
outputData = removePlaceholderSpeech(outputData, "選択:「１．", "\n:１．")
outputData = reduceWhitespaceBeforeNumbers(outputData)
outputData = addColonBeforeDecisions(outputData)
outputData = replaceSpecialCharacters(outputData)
outputData = utils.removeQuoteMarksFromSpeech(outputData)
outputData = utils.simplifyLongNewlineSequences(outputData)
outputData = utils.addLeadingWhitespaceWhenSameSpeaker(
outputData, True)
outputData = utils.addSpaceAfterChoiceMarker(outputData)
with io.open(os.path.join(outputPath, fn), "w",
encoding="utf-8") as fd:
fd.write(outputData)
def extractDialogAsJson(rootDir, outputFn, builder, title, version):
sectionNum = 0
sectionsByFile = []
for fn, fd in utils.loadFiles(rootDir):
sections, sectionNum = compileSections(fd, builder, sectionNum)
sectionsByFile.append({"source": fn, "sections": sections})
root = {"game_content": title, "version": version, "content": sectionsByFile}
with io.open(outputFn, "w", encoding="utf-8") as fd:
json.dump(root, fd, ensure_ascii=False)
def clean(root, outputPath):
"""
Post-clean up thoughts
This has a lot going on. Changes in the scene are demarcated with < >.
Keywords, sometimes the current speaker, and maybe narrations are decorated
with 【 】
A blank line represents a break in continuity or passage of time maybe?
All dialog can be extracted by picking lines with ":" in them.
"""
if not os.path.exists(outputPath):
os.mkdir(outputPath)
for fn, fd in utils.loadFiles(root):
data = fd.read()
data = data.lstrip()
data = data[data.find("\n"):].strip()
start = 0
while True:
start = data.find("「", start)
if start == -1:
break
end = data.find("」", start)
speech = data[start:end].replace("\n", "")
data = data[:start] + speech + data[end:]
start = start + 1
data = data.replace("\n「", ":「")
data = data.replace("」:「", "」\n:「")
data = data.split("TO BE")[0]
data = data.strip() + "\n"
data = changeEventMarking(data)
data = removeCharactersSurroundingSpeaker(data)
data = utils.removeQuoteMarksFromSpeech(data)
data = utils.simplifyLongNewlineSequences(data)
data = utils.addSpaceAfterChoiceMarker(data)
data = utils.addLeadingWhitespaceWhenSameSpeaker(data, True)
with io.open(os.path.join(outputPath, fn), "w",
encoding="utf-8") as fd:
fd.write(data)
def clean(root, outputPath):
"""
Post-clean up thoughts
-- had I parsed the page as a table, rather than do a text dump,
the process would have been simpler. I had to manually
pre-process the last page (credits). Otherwise, some pages have
a table with 2 columns and others have 3 columns.
"""
if not os.path.exists(outputPath):
os.mkdir(outputPath)
for fn, fd in utils.loadFiles(root):
data = fd.read()
data = data.lstrip()
pageTitle, data = data.split("\n", 1)
dataList = data.split("セリフ＆ナレーション")
if len(dataList) != 2:
continue
data = dataList[1]
data = data.split("Back Next")[0].strip()
data = data.replace("\n ", " ")
step = 2
if data.count("\n \n") > 5:
step = 3
outputRows = [f": ＋＋{pageTitle}"]
dataList = data.split("\n")
for i in range(0, len(dataList) - step + 1, step):
speaker = dataList[i]
text = dataList[i + step - 1]
outputRows.append("%s:「%s」" % (speaker, text))
data = "\n".join(outputRows) + "\n"
data = utils.removeQuoteMarksFromSpeech(data)
data = utils.simplifyLongNewlineSequences(data)
data = utils.addSpaceAfterChoiceMarker(data)
data = utils.removeReundantSpeakers(data)
data = utils.addLeadingWhitespaceWhenSameSpeaker(data, True)
with io.open(os.path.join(outputPath, fn), "w",
encoding="utf-8") as fd:
fd.write(data)
def extractDialogWithMecab(rootDir):
outputDir = utils.getOutputPath(rootDir, "stats")
parser = MeCab.Tagger("-Owakati")
unigrams = []
bigrams = []
trigrams = []
fourgrams = []
for fn, fd in utils.loadFiles(rootDir):
for line in fd:
wordList = parser.parse(line).split()
unigrams.extend(getChunks(wordList, 1))
bigrams.extend(getChunks(wordList, 2))
trigrams.extend(getChunks(wordList, 3))
fourgrams.extend(getChunks(wordList, 4))
_output(outputDir, unigrams, bigrams, trigrams, fourgrams)
def extractDialogAsHtml(rootDir,
outputDir,
builder,
tocTitle,
getTitleFromFile=None):
if not os.path.exists(outputDir):
os.mkdir(outputDir)
sectionNum = 0
for fn, fd in utils.loadFiles(rootDir):
sections, sectionNum = compileSections(fd, builder, sectionNum)
sectionsText = "\n\n".join(sections)
html = builder.makeDocument(sectionsText)
outputFn = os.path.splitext(fn)[0] + ".html"
with io.open(os.path.join(outputDir, outputFn), "w",
encoding="utf-8") as fd:
fd.write(html)
generateDirectoryListing(outputDir, tocTitle, getTitleFromFile)
def clean(sourceFolder, outputFolder):
if not os.path.exists(outputFolder):
os.mkdir(outputFolder)
for fn, fd in utils.loadFiles(sourceFolder):
section = fd.read()
section = condenseWhitespace(section)
section = removeNewlinesInContinuousSpeech(section)
section = markEvents(section)
section = separateSpeakersAndSpeech(section)
section = section.replace(" ", "")
section = section.strip()
section = utils.simplifyLongNewlineSequences(section)
section = utils.addSpaceAfterChoiceMarker(section)
section = utils.addLeadingWhitespaceWhenSameSpeaker(section, False)
with io.open(join(outputFolder, fn), "w", encoding="utf-8") as writeFd:
writeFd.write(section)
def clean(root, outputPath):
"""
Post-clean up thoughts
Seems similar to FF7 but with more whitespace.
Lots of inner monologue from Squall?
"""
locationRe = re.compile("\n\n[^\n(（『「①②③＋]+\n\n")
if not os.path.exists(outputPath):
os.mkdir(outputPath)
for fn, fd in utils.loadFiles(root):
data = fd.read()
data = data.strip()
# For unmarked locations, mark them
start = 0
while True:
match = locationRe.search(data, start)
if match is None:
break
start = match.start() + 2 # pass the 2 leading newlines
# text = data[match.start():match.end()]
# if text.strip() == '':
# continue
data = data[:start] + "＋＋" + data[start:]
start += 2 # pass the 2 inserted characters
data = data.replace("\n（", ":（")
# Place quoted sections onto a single line
start = 0
while True:
start = data.find("「", start)
if start == -1:
break
end = data.find("」", start)
speech = data[start:end].replace("\n", "")
data = data[:start] + speech + data[end:]
start = start + 1
data = data.replace("\n「", ":「")
data = data.replace("」:「", "」\n:「")
data = data.replace("\n\n", "\n")
data = markChoices(data)
data = correctNumberOfQuotes(data)
data = utils.removeQuoteMarksFromSpeech(data)
data = utils.simplifyLongNewlineSequences(data)
data = utils.addSpaceAfterChoiceMarker(data)
data = utils.addLeadingWhitespaceWhenSameSpeaker(data, False)
data = data.split("＊＊＊＊＊")[0]
data = data.split("＋＋私的好き台詞＋＋")[0]
data = data.strip() + "\n"
data = "＋＋" + data
data = data.replace("\n", "＋＋\n", 1)
with io.open(os.path.join(outputPath, fn), "w", encoding="utf-8") as fd:
fd.write(data)
