def load_annotation(gold_file):
source_sentences = []
gold_edits = []
fgold = smart_open(gold_file, 'r')
puffer = fgold.read()
fgold.close()
puffer = puffer.decode('utf8')
for item in paragraphs(puffer.splitlines(True)):
item = item.splitlines(False)
sentence = [line[2:].strip() for line in item if line.startswith('S ')]
assert sentence != []
annotation = []
for line in item[1:]:
if line.startswith('I ') or line.startswith('S '):
continue
assert line.startswith('A ')
line = line[2:]
fields = line.split('|||')
start_offset = int(fields[0].split()[0])
end_offset = int(fields[0].split()[1])
etype = fields[1]
corrections = [c.strip() if c != '-NONE-' else '' for c in fields[2].split('||')]
# NOTE: start and end are *token* offsets
original = ' '.join(' '.join(sentence).split()[start_offset:end_offset])
annotation.append((start_offset, end_offset, original, corrections))
tok_offset = 0
for this_sentence in sentence:
tok_offset += len(this_sentence.split())
this_edits = [edit for edit in annotation if edit[0] <= tok_offset and edit[1] <= tok_offset]
source_sentences.append(this_sentence)
gold_edits.append(this_edits)
return (source_sentences, gold_edits)
def load_annotation(gold_file):
source_sentences = []
gold_edits = []
fgold = smart_open(gold_file, 'r')
puffer = fgold.read()
fgold.close()
#puffer = puffer.decode('utf8')
for item in paragraphs(puffer.splitlines(True)):
item = item.splitlines(False)
sentence = [line[2:].strip() for line in item if line.startswith('S ')]
#print(sentence)
assert sentence != []
annotations = {}
for line in item[1:]:
if line.startswith('I ') or line.startswith('S '):
continue
assert line.startswith('A ')
line = line[2:]
fields = line.split('|||')
start_offset = int(fields[0].split()[0])
end_offset = int(fields[0].split()[1])
etype = fields[1]
if etype == 'noop':
start_offset = -1
end_offset = -1
corrections = [
c.strip() if c != '-NONE-' else ''
for c in fields[2].split('||')
]
# NOTE: start and end are *token* offsets
original = ' '.join(
' '.join(sentence).split()[start_offset:end_offset])
annotator = int(fields[5])
if annotator not in list(annotations.keys()):
annotations[annotator] = []
# print(etype, original,corrections)
annotations[annotator].append(
(start_offset, end_offset, original, corrections, etype))
tok_offset = 0
for this_sentence in sentence:
tok_offset += len(this_sentence.split())
source_sentences.append(this_sentence)
this_edits = {}
for annotator, annotation in annotations.items():
this_edits[annotator] = [
edit for edit in annotation if edit[0] <= tok_offset
and edit[1] <= tok_offset and edit[0] >= 0 and edit[1] >= 0
]
if len(this_edits) == 0:
this_edits[0] = []
#print(this_edits)
gold_edits.append(this_edits)
return (source_sentences, gold_edits)
def load_annotation(gold_file):
source_sentences = []
gold_edits = []
fgold = smart_open(gold_file, "r")
puffer = fgold.read()
fgold.close()
puffer = puffer.decode("utf8")
for item in paragraphs(puffer.splitlines(True)):
item = item.splitlines(False)
sentence = [line[2:].strip() for line in item if line.startswith("S ")]
assert sentence != []
annotations = {}
for line in item[1:]:
if line.startswith("I ") or line.startswith("S "):
continue
assert line.startswith("A ")
line = line[2:]
fields = line.split("|||")
start_offset = int(fields[0].split()[0])
end_offset = int(fields[0].split()[1])
etype = fields[1]
if etype == "noop":
start_offset = -1
end_offset = -1
corrections = [c.strip() if c != "-NONE-" else "" for c in fields[2].split("||")]
# NOTE: start and end are *token* offsets
original = " ".join(" ".join(sentence).split()[start_offset:end_offset])
annotator = int(fields[5])
if annotator not in annotations.keys():
annotations[annotator] = []
annotations[annotator].append((start_offset, end_offset, original, corrections))
tok_offset = 0
for this_sentence in sentence:
tok_offset += len(this_sentence.split())
source_sentences.append(this_sentence)
this_edits = {}
for annotator, annotation in annotations.iteritems():
this_edits[annotator] = [
edit
for edit in annotation
if edit[0] <= tok_offset and edit[1] <= tok_offset and edit[0] >= 0 and edit[1] >= 0
]
if len(this_edits) == 0:
this_edits[0] = []
gold_edits.append(this_edits)
return (source_sentences, gold_edits)
def load_annotation(gold_file):
source_sentences = []
gold_edits = []
fgold = smart_open(gold_file, 'r')
puffer = fgold.read()
fgold.close()
puffer = puffer.decode('utf8')
for item in paragraphs(puffer.splitlines(True)):
item = item.splitlines(False)
sentence = [line[2:].strip() for line in item if line.startswith('S ')]
assert sentence != []
annotation = []
for line in item[1:]:
if line.startswith('I ') or line.startswith('S '):
continue
assert line.startswith('A ')
line = line[2:]
fields = line.split('|||')
start_offset = int(fields[0].split()[0])
end_offset = int(fields[0].split()[1])
etype = fields[1]
corrections = [
c.strip() if c != '-NONE-' else ''
for c in fields[2].split('||')
]
# NOTE: start and end are *token* offsets
original = ' '.join(
' '.join(sentence).split()[start_offset:end_offset])
annotation.append(
(start_offset, end_offset, original, corrections))
tok_offset = 0
for this_sentence in sentence:
tok_offset += len(this_sentence.split())
this_edits = [
edit for edit in annotation
if edit[0] <= tok_offset and edit[1] <= tok_offset
]
source_sentences.append(this_sentence)
gold_edits.append(this_edits)
return (source_sentences, gold_edits)
opts, args = getopt.getopt(sys.argv[1:], "", ["out_train=", "out_dev=", "out_test="])
train_out = 'train'
dev_out = 'dev'
test_out = 'test'
for o, v in opts:
if o == "--out_train":
train_out = v
elif o == "--out_dev":
dev_out = v
elif o == "--out_test":
test_out = v
else:
print >> sys.stderr, "Unknown option : ", o
assert len(args) == 2
data = list(paragraphs(sys.stdin))
total = len(data)
train_split, dev_split = map(lambda i : int(i) * total / 100, args)
test_split = total - train_split - dev_split
data_iter = iter(data)
for no_instances, output in zip([train_split, dev_split, test_split], [train_out, dev_out, test_out]):
fout = open(output, 'wb')
for i in xrange(no_instances):
fout.write(next(data_iter) + '\n')
def load_annotation(gold_file):
source_sentences = []
gold_edits = []
fgold = smart_open(gold_file, 'r')
puffer = fgold.read()
fgold.close()
puffer = puffer.decode('utf8')
for item in paragraphs(puffer.splitlines(True)):
item = item.splitlines(False)
sentence = [line[2:].strip() for line in item if line.startswith('S ')]
assert sentence != []
annotations = {}
for line in item[1:]:
if line.startswith('I ') or line.startswith('S '):
continue
assert line.startswith('A ')
line = line[2:]
fields = line.split('|||')
start_offset = int(fields[0].split()[0])
end_offset = int(fields[0].split()[1])
etype = fields[1]
if etype == 'noop':
start_offset = -1
end_offset = -1
corrections = [
c.strip() if c != '-NONE-' else ''
for c in fields[2].split('||')
]
# NOTE: start and end are *token* offsets
original = ' '.join(
' '.join(sentence).split()[start_offset:end_offset])
annotator = int(fields[5])
if annotator not in annotations.keys():
annotations[annotator] = []
annotations[annotator].append(
(start_offset, end_offset, original, corrections))
tok_offset = 0
for this_sentence in sentence:
tok_offset += len(this_sentence.split())
source_sentences.append(this_sentence)
this_edits = {}
for annotator, annotation in annotations.iteritems():
this_edits[annotator] = [
edit for edit in annotation if edit[0] <= tok_offset
and edit[1] <= tok_offset and edit[0] >= 0 and edit[1] >= 0
]
if len(this_edits) == 0:
this_edits[0] = []
gold_edits.append(this_edits)
return (source_sentences, gold_edits)
print >> sys.stderr, "Usage: m2scorer.py [OPTIONS] proposed_sentences gold_source"
print >> sys.stderr, "where"
print >> sys.stderr, " proposed_sentences - system output, sentence per line"
print >> sys.stderr, " source_gold - source sentences with gold token edits"
print >> sys.stderr, "OPTIONS"
print >> sys.stderr, " -v --verbose - print verbose output"
print >> sys.stderr, " --very_verbose - print lots of verbose output"
print >> sys.stderr, " --max_unchanged_words N - Maximum unchanged words when extraction edit. Default 2."
print >> sys.stderr, " --beta B - Beta value for F-measure. Default 0.5."
print >> sys.stderr, " --ignore_whitespace_casing - Ignore edits that only affect whitespace and caseing. Default no."
import getopt
opts, args = getopt.getopt(sys.argv[1:], "",
["out_train=", "out_dev=", "out_test="])
train_out = 'train'
dev_out = 'dev'
test_out = 'test'
for o, v in opts:
if o == "--out_train":
train_out = v
elif o == "--out_dev":
dev_out = v
elif o == "--out_test":
test_out = v
else:
print >> sys.stderr, "Unknown option : ", o
assert len(args) == 2
data = list(paragraphs(sys.stdin))
total = len(data)
train_split, dev_split = map(lambda i: int(i) * total / 100, args)
test_split = total - train_split - dev_split
data_iter = iter(data)
for no_instances, output in zip([train_split, dev_split, test_split],
[train_out, dev_out, test_out]):
fout = open(output, 'wb')
for i in xrange(no_instances):
fout.write(next(data_iter) + '\n')
