class Evaluator:
def __init__(self, node, google_key=None, use_sessions=True):
self._engine = node.engine
self._node = node
self._heval_outputter = HumanEvaluationFileOutputter()
self._xmlencoder = XMLEncoder()
self._translators = [
GoogleTranslate(self._engine.source_lang, self._engine.target_lang, key=google_key),
# BingTranslator(source_lang, target_lang),
MMTTranslator(self._node, use_sessions)
]
def evaluate(self, corpora, heval_output=None, debug=False):
if len(corpora) == 0:
raise IllegalArgumentException('empty corpora')
if heval_output is not None:
fileutils.makedirs(heval_output, exist_ok=True)
target_lang = self._engine.target_lang
source_lang = self._engine.source_lang
logger = _evaluate_logger()
logger.start(corpora)
working_dir = self._engine.get_tempdir('evaluation')
try:
results = []
# Process references
with logger.step('Preparing corpora') as _:
corpora_path = os.path.join(working_dir, 'corpora')
corpora = self._xmlencoder.encode(corpora, corpora_path)
reference = os.path.join(working_dir, 'reference.' + target_lang)
source = os.path.join(working_dir, 'source.' + source_lang)
fileutils.merge([corpus.get_file(target_lang) for corpus in corpora], reference)
fileutils.merge([corpus.get_file(source_lang) for corpus in corpora], source)
if heval_output is not None:
self._heval_outputter.write(lang=target_lang, input_file=reference,
output_file=os.path.join(heval_output, 'reference.' + target_lang))
self._heval_outputter.write(lang=source_lang, input_file=source,
output_file=os.path.join(heval_output, 'source.' + source_lang))
# Translate
for translator in self._translators:
name = translator.name()
with logger.step('Translating with %s' % name) as _:
result = _EvaluationResult(translator)
results.append(result)
translations_path = os.path.join(working_dir, 'translations', result.id + '.raw')
xmltranslations_path = os.path.join(working_dir, 'translations', result.id)
fileutils.makedirs(translations_path, exist_ok=True)
try:
translated, mtt, parallelism = translator.translate(corpora, translations_path)
filename = result.id + '.' + target_lang
result.mtt = mtt
result.parallelism = parallelism
result.translated_corpora = self._xmlencoder.encode(translated, xmltranslations_path)
result.merge = os.path.join(working_dir, filename)
fileutils.merge([corpus.get_file(target_lang)
for corpus in result.translated_corpora], result.merge)
if heval_output is not None:
self._heval_outputter.write(lang=target_lang, input_file=result.merge,
output_file=os.path.join(heval_output, filename))
except TranslateError as e:
result.error = e
except Exception as e:
result.error = TranslateError('Unexpected ERROR: ' + str(e.message))
# Check corpora length
reference_lines = fileutils.linecount(reference)
for result in results:
if result.error is not None:
continue
lines = fileutils.linecount(result.merge)
if lines != reference_lines:
raise TranslateError('Invalid line count for translator %s: expected %d, found %d.'
% (result.translator.name(), reference_lines, lines))
# Scoring
scorers = [(MatecatScore(), 'pes'), (BLEUScore(), 'bleu')]
for scorer, field in scorers:
with logger.step('Calculating %s' % scorer.name()) as _:
for result in results:
if result.error is not None:
continue
setattr(result, field, scorer.calculate(result.merge, reference))
logger.completed(results, scorers)
return results
finally:
if not debug:
self._engine.clear_tempdir('evaluation')
class Evaluator:
def __init__(self, node, google_key=None, google_nmt=False):
self._engine = node.engine
self._node = node
self._heval_outputter = HumanEvaluationFileOutputter()
self._xmlencoder = XMLEncoder()
self._translators = [
GoogleTranslate(self._engine.source_lang, self._engine.target_lang, key=google_key, nmt=google_nmt),
# BingTranslator(source_lang, target_lang),
MMTTranslator(self._node)
]
def evaluate(self, corpora, heval_output=None, debug=False):
target_lang = self._engine.target_lang
source_lang = self._engine.source_lang
corpora = [corpus for corpus in corpora if source_lang in corpus.langs and target_lang in corpus.langs]
if len(corpora) == 0:
raise IllegalArgumentException('No %s > %s corpora found into specified path' % (source_lang, target_lang))
if heval_output is not None:
fileutils.makedirs(heval_output, exist_ok=True)
logger = _evaluate_logger()
logger.start(corpora)
working_dir = self._engine.get_tempdir('evaluation')
try:
results = []
# Process references
with logger.step('Preparing corpora') as _:
corpora_path = os.path.join(working_dir, 'corpora')
corpora = self._xmlencoder.encode(corpora, corpora_path)
reference = os.path.join(working_dir, 'reference.' + target_lang)
source = os.path.join(working_dir, 'source.' + source_lang)
fileutils.merge([corpus.get_file(target_lang) for corpus in corpora], reference)
fileutils.merge([corpus.get_file(source_lang) for corpus in corpora], source)
if heval_output is not None:
self._heval_outputter.write(lang=target_lang, input_file=reference,
output_file=os.path.join(heval_output, 'reference.' + target_lang))
self._heval_outputter.write(lang=source_lang, input_file=source,
output_file=os.path.join(heval_output, 'source.' + source_lang))
# Translate
for translator in self._translators:
name = translator.name()
with logger.step('Translating with %s' % name) as _:
result = _EvaluationResult(translator)
results.append(result)
translations_path = os.path.join(working_dir, 'translations', result.id + '.raw')
xmltranslations_path = os.path.join(working_dir, 'translations', result.id)
fileutils.makedirs(translations_path, exist_ok=True)
try:
translated, mtt, parallelism = translator.translate(corpora, translations_path)
filename = result.id + '.' + target_lang
result.mtt = mtt
result.parallelism = parallelism
result.translated_corpora = self._xmlencoder.encode(translated, xmltranslations_path)
result.merge = os.path.join(working_dir, filename)
fileutils.merge([corpus.get_file(target_lang)
for corpus in result.translated_corpora], result.merge)
if heval_output is not None:
self._heval_outputter.write(lang=target_lang, input_file=result.merge,
output_file=os.path.join(heval_output, filename))
except TranslateError as e:
result.error = e
except Exception as e:
result.error = TranslateError('Unexpected ERROR: ' + str(e.message))
# Check corpora length
reference_lines = fileutils.linecount(reference)
for result in results:
if result.error is not None:
continue
lines = fileutils.linecount(result.merge)
if lines != reference_lines:
raise TranslateError('Invalid line count for translator %s: expected %d, found %d.'
% (result.translator.name(), reference_lines, lines))
# Scoring
scorers = [(MatecatScore(), 'pes'), (BLEUScore(), 'bleu')]
for scorer, field in scorers:
with logger.step('Calculating %s' % scorer.name()) as _:
for result in results:
if result.error is not None:
continue
setattr(result, field, scorer.calculate(result.merge, reference))
logger.completed(results, scorers)
return results
finally:
if not debug:
self._engine.clear_tempdir('evaluation')
class BatchTranslator:
def __init__(self, node, use_sessions=True):
self._engine = node.engine
self._node = node
self._xmlencoder = XMLEncoder()
self._translator = MMTTranslator(self._node, use_sessions)
def translate(self, corpora, dest_path=None, debug=False):
if len(corpora) == 0:
raise IllegalArgumentException('empty corpora')
if dest_path:
fileutils.makedirs(dest_path, exist_ok=True)
target_lang = self._engine.target_lang
source_lang = self._engine.source_lang
working_dir = self._engine.get_tempdir('evaluation')
have_references = False
try:
results = []
# Process references
corpora_path = os.path.join(working_dir, 'corpora')
corpora = self._xmlencoder.encode(corpora, corpora_path)
reference = os.path.join(working_dir, 'reference.' + target_lang)
source = os.path.join(working_dir, 'source.' + source_lang)
refs = [corpus.get_file(target_lang) for corpus in corpora if corpus.get_file(target_lang)]
have_references = len(refs) > 0
fileutils.merge(refs, reference) # tolerates missing reference
fileutils.merge([corpus.get_file(source_lang) for corpus in corpora], source)
if dest_path:
for corpus in corpora:
corpus.copy(dest_path, suffixes={source_lang: '.src', target_lang: '.ref', 'tmx': '.src'})
# Translate
translator = self._translator
name = translator.name()
result = _EvaluationResult(translator)
results.append(result)
translations_path = os.path.join(working_dir, 'translations', result.id + '.raw')
xmltranslations_path = os.path.join(working_dir, 'translations', result.id)
fileutils.makedirs(translations_path, exist_ok=True)
try:
translated, mtt, parallelism = translator.translate(corpora, translations_path)
filename = result.id + '.' + target_lang
result.mtt = mtt
result.parallelism = parallelism
result.translated_corpora = self._xmlencoder.encode(translated, xmltranslations_path)
result.merge = os.path.join(working_dir, filename)
fileutils.merge([corpus.get_file(target_lang)
for corpus in result.translated_corpora], result.merge)
if dest_path:
for corpus in result.translated_corpora:
corpus.copy(dest_path, suffixes={target_lang: '.hyp', 'tmx': '.hyp'})
except TranslateError as e:
result.error = e
except Exception as e:
result.error = TranslateError('Unexpected ERROR: ' + str(e.message))
if result.error is None:
if have_references:
scorer = BLEUScore()
# bleu in range [0;1)
bleu = scorer.calculate(result.merge, reference)
return bleu
else:
return True
else:
print(result.error)
return None
finally:
if not debug:
self._engine.clear_tempdir('evaluation')