def main():
"""Main method for computing Oracle WER."""
parser = argparse.ArgumentParser()
parser.add_argument(
"nbest_file",
type=open_file_stream,
help=
'A file containing n-best lists. Read as a gzip file if filename ends with .gz'
)
parser.add_argument("ref_file", type=argparse.FileType('r'))
parser.add_argument("output", type=argparse.FileType('w'))
args = parser.parse_args()
print('Reading n-best lists...')
nbests = list(read_nbest_file(args.nbest_file))
print('# of nbests: {}'.format(len(nbests)))
print('Reading transcripts...')
refs = read_transcript_table(args.ref_file)
asr_tools.evaluation_util.REFERENCES = refs
# This is the slow part.
print('Running evaluation...')
overall_eval = evaluate_nbests(nbests)
# Write them back out to a file
write_nbests(args.output, nbests, save_eval=True)
args.nbest_file.close()
def main():
"""Main method for computing Oracle WER."""
parser = argparse.ArgumentParser()
parser.add_argument("nbest_file", type=open_file_stream, help='A file containing n-best lists. Read as a gzip file if filename ends with .gz')
parser.add_argument("ref_file", type=argparse.FileType('r'))
# parser.add_argument('--plot', '-p', default=False, action='store_true')
args = parser.parse_args()
print('Reading n-best lists...')
nbests = list(read_nbest_file(args.nbest_file))
print('# of nbests: {}'.format(len(nbests)))
print('Reading transcripts...')
refs = read_transcript_table(args.ref_file)
asr_tools.evaluation_util.REFERENCES = refs
# This is the slow part.
print('Running evaluation...')
overall_eval = evaluate_nbests(nbests)
print('Overall eval:')
print(overall_eval)
print()
print('Computing oracle eval...')
print('Oracle eval:')
print(evaluate_nbests_oracle(nbests))
evals = evals_by_depth(nbests)
wers = list(map(lambda x: x.wer(), evals))
# if args.plot:
# plt.plot(wers)
# plt.ylim(ymin=0)
# plt.show()
args.nbest_file.close()
def setUpClass(cls):
# Load the major objects that we'll use for the tests
with open(cls.ref_file) as f:
cls.refs = read_transcript_table(f)
with open(cls.ref_file) as f:
set_global_references(f)
with open(cls.hyp_file) as f:
cls.hyps = read_transcript(f)
with open(cls.nbest_file) as f:
cls.nbests = list(read_nbest_file(f))
cls.s1 = cls.hyps[0]
cls.s2 = get_global_reference(cls.s1.id_)
# Evaluate each and check their WER
cls.e1 = evaluate(cls.refs, cls.s1)
cls.e2 = evaluate(cls.refs, cls.s2)
def setUpClass(cls):
# Load the major objects that we'll use for the tests
with open(cls.ref_file) as f:
cls.refs = read_transcript_table(f)
with open(cls.ref_file) as f:
set_global_references(f)
with open(cls.hyp_file) as f:
cls.hyps = read_transcript(f)
with open(cls.nbest_file) as f:
cls.nbests = list(read_nbest_file(f))
cls.s1 = cls.hyps[0]
cls.s2 = get_global_reference(cls.s1.id_)
# Evaluate each and check their WER
cls.e1 = evaluate(cls.refs, cls.s1)
cls.e2 = evaluate(cls.refs, cls.s2)
def main():
"""Main method for figuring out which examples from n-best lists
are potentially improvable."""
parser = argparse.ArgumentParser()
parser.add_argument("nbest_file", type=open_file_stream, help='A file containing n-best lists. Read as a gzip file if filename ends with .gz')
parser.add_argument("ref_file", type=argparse.FileType('r'))
args = parser.parse_args()
nbests = list(read_nbest_file(args.nbest_file))
refs = read_transcript_table(args.ref_file)
asr_tools.evaluation_util.REFERENCES = refs
overall_eval = evaluate_nbests(nbests)
for nbest in nbests:
if nbest.is_improveable():
print_nbest_ref_hyp_best(nbest)
# print_nbest(nbest)
print(overall_eval)
def main():
"""Main method to show n-best lists, printing to console."""
parser = argparse.ArgumentParser()
parser.add_argument("nbest_file", type=argparse.FileType('r'))
parser.add_argument("ref_file", nargs='?', type=argparse.FileType('r')) # optional
parser.add_argument("--verbose", '-v', default=True)
args = parser.parse_args()
colorama.init()
nbests = list(read_nbest_file(args.nbest_file))
if args.ref_file:
refs = read_transcript_table(args.ref_file)
asr_tools.evaluation_util.REFERENCES = refs
overall_eval = evaluate_nbests(nbests)
for nbest in nbests:
print('NBEST:')
print_nbest(nbest, acscore=True, lmscore=True, tscore=True, maxwords=10)
if not monotone(nbest.sentences, comparison=operator.lt, key=Sentence.score):
print(termcolor.colored('WARNING: Non-montonic scores', 'red', attrs=['bold']))
if args.ref_file:
print(overall_eval)
def main():
"""Main method for computing Oracle WER."""
parser = argparse.ArgumentParser()
parser.add_argument(
"nbest_file",
type=open_file_stream,
help=
'A file containing n-best lists. Read as a gzip file if filename ends with .gz'
)
parser.add_argument("ref_file", type=argparse.FileType('r'))
# parser.add_argument('--plot', '-p', default=False, action='store_true')
args = parser.parse_args()
print('Reading n-best lists...')
nbests = list(read_nbest_file(args.nbest_file))
print('# of nbests: {}'.format(len(nbests)))
print('Reading transcripts...')
refs = read_transcript_table(args.ref_file)
asr_tools.evaluation_util.REFERENCES = refs
# This is the slow part.
print('Running evaluation...')
overall_eval = evaluate_nbests(nbests)
print('Overall eval:')
print(overall_eval)
print()
print('Computing oracle eval...')
print('Oracle eval:')
print(evaluate_nbests_oracle(nbests))
evals = evals_by_depth(nbests)
wers = list(map(lambda x: x.wer(), evals))
# if args.plot:
# plt.plot(wers)
# plt.ylim(ymin=0)
# plt.show()
args.nbest_file.close()
def main():
"""Main method for figuring out which examples from n-best lists
are potentially improvable."""
parser = argparse.ArgumentParser()
parser.add_argument(
"nbest_file",
type=open_file_stream,
help=
'A file containing n-best lists. Read as a gzip file if filename ends with .gz'
)
parser.add_argument("ref_file", type=argparse.FileType('r'))
args = parser.parse_args()
nbests = list(read_nbest_file(args.nbest_file))
refs = read_transcript_table(args.ref_file)
asr_tools.evaluation_util.REFERENCES = refs
overall_eval = evaluate_nbests(nbests)
for nbest in nbests:
if nbest.is_improveable():
print_nbest_ref_hyp_best(nbest)
# print_nbest(nbest)
print(overall_eval)
def main():
"""Main method for computing Oracle WER."""
parser = argparse.ArgumentParser()
parser.add_argument("nbest_file", type=open_file_stream, help='A file containing n-best lists. Read as a gzip file if filename ends with .gz')
parser.add_argument("ref_file", type=argparse.FileType('r'))
parser.add_argument("output", type=argparse.FileType('w'))
args = parser.parse_args()
print('Reading n-best lists...')
nbests = list(read_nbest_file(args.nbest_file))
print('# of nbests: {}'.format(len(nbests)))
print('Reading transcripts...')
refs = read_transcript_table(args.ref_file)
asr_tools.evaluation_util.REFERENCES = refs
# This is the slow part.
print('Running evaluation...')
overall_eval = evaluate_nbests(nbests)
# Write them back out to a file
write_nbests(args.output, nbests, save_eval=True)
args.nbest_file.close()
def test_read_nbest(self):
# Make sure we can read an n-best and get the right number of nbest lists back
with open(self.nbest_file) as f:
nbests = list(read_nbest_file(f))
self.assertTrue(len(nbests) == 15)
def test_read_nbest(self):
# Make sure we can read an n-best and get the right number of nbest lists back
with open(self.nbest_file) as f:
nbests = list(read_nbest_file(f))
self.assertTrue(len(nbests) == 15)