references = [line.strip().lower() for line in trg_file]
else:
hypotheses = [line.strip() for line in src_file]
references = [line.strip() for line in trg_file]
if args.max_size is not None:
hypotheses = hypotheses[:args.max_size]
references = references[:args.max_size]
if len(hypotheses) != len(references):
sys.stderr.write(
'warning: source and target don\'t have the same length\n')
size = min(len(hypotheses), len(references))
hypotheses = hypotheses[:size]
references = references[:size]
avg_stats, stats = tercom_statistics(hypotheses, references)
ters = [stats_['TER'] for stats_ in stats]
mean = sum(ters) / len(ters)
variance = sum((ter - mean)**2 for ter in ters) / (len(ters) - 1)
ts = {0.01: 2.5841, 0.05: 1.9639, 0.10: 1.6474}
t = ts.get(args.p)
if t is None:
raise Exception
d = t * np.sqrt(variance / len(ters))
print('{:.3f} +/- {:.3f}'.format(mean, d))
parser.add_argument('--average', nargs='+', type=int)
if __name__ == '__main__':
args = parser.parse_args()
with open(args.ref_file) as f:
references = [line.strip() for line in f]
hypotheses = []
for hyp_file in args.hyp_files:
with open(hyp_file) as f:
hypotheses.append([line.strip() for line in f])
if args.reverse:
scores = [tercom_statistics(references, hyp)[0] for hyp in hypotheses]
else:
scores = [tercom_statistics(hyp, references)[0] for hyp in hypotheses]
N = len(args.average) if args.average else len(args.hyp_files)
ind = np.arange(N)
op_name_mapping = {'ins': 'Insertions', 'del': 'Deletions', 'sub': 'Substitutions', 'shift': 'Shifts'}
ref_words = np.array([score["REF_WORDS"] for score in scores])
bars = []
legend = []
bottom = np.zeros(N)
colors = ['#e66101', '#fdb863', '#b2abd2', '#5e3c99']
n = 1000
avg_length = 0
while True:
i += 1
hypotheses = list(islice(src_file, n))
references = list(islice(trg_file, n))
if not hypotheses or not references:
break
hypotheses = [line.strip() for line in hypotheses]
references = [line.strip() for line in references]
_, stats = tercom_statistics(hypotheses, references, not args.case_insensitive)
if avg_length == 0:
avg_length = sum(stats_['REF_WORDS'] for stats_ in stats) / len(stats)
for stats_ in stats:
for field in ('DEL', 'INS', 'SUB', 'WORD_SHIFT'):
stats_[field] /= stats_['REF_WORDS']
stats_['REF_WORDS'] = (stats_['REF_WORDS'] - avg_length) / avg_length
stats_['TER'] /= 100
if not args.output:
print('\n'.join(','.join(str(round(stats_[k], args.precision)) for k in fields)
for stats_ in stats))
else: