def evaluate_block(self, data_iter: DataLoader,
docs: Sequence[Document]) -> float:
"""
:param data_iter:
:param docs:
:return:
"""
self.decode_block(data_iter=data_iter, docs=docs)
if self.chunking:
acc = ChunkF1()
for doc in docs:
for sen in doc.sentences:
acc.update(labels=sen[to_gold(self.key)],
preds=sen[self.key])
else:
acc = Accuracy()
for doc in docs:
for sen in doc.sentences:
labels = nd.array([
self.label_map.cid(label)
for label in sen[to_gold(self.key)]
])
preds = nd.array(
[self.label_map.cid(pred) for pred in sen[self.key]])
acc.update(labels=labels, preds=preds)
return acc.get()[1]
def json_reader(filepath: str,
cols: Any = None,
key: str = None) -> Tuple[List[Document], LabelMap]:
# TODO: update this to accept any format (see tsv_reader)
documents = []
dc = wc = sc = 0
label_map = LabelMap()
logging.info('Reading json file from: ')
if not os.path.isfile(filepath) and filepath.endswith('.json'):
raise ValueError("{} is not a valid format".format(filepath))
logging.info('- filepath: %s' % filepath)
with open(filepath) as f:
docs = json.load(f)
for i, doc in enumerate(docs):
sentences = []
for sen in doc['sens']:
wc += len(sen['tok'])
if key is not None:
sen = sen.copy()
sen[to_gold(key)] = sen.pop(key)
sentences.append(Sentence(sen))
sc += len(sentences)
[[label_map.add(i) for i in sent[to_gold(key)]] for sent in sentences]
document = Document(sens=sentences)
document[DOC_ID] = i
documents.append(document)
dc += len(documents)
logging.info('- dc = %d, sc = %d, wc = %d' % (dc, sc, wc))
return documents, label_map
def tsv_reader(tsv_directory: str,
cols: Dict[str, int],
key: str = None) -> Tuple[List[Document], LabelMap]:
documents = []
wc = sc = 0
label_map = LabelMap()
if TOK not in cols:
raise ValueError('The column index of "%s" must be specified' % TOK)
if key is not None:
if key in cols:
cols = cols.copy()
cols[to_gold(key)] = cols.pop(key)
else:
raise ValueError('Key mismatch: %s is not a key in %s' % (key, str(cols)))
logging.info('Reading tsv from:')
logging.info('- directory: %s' % tsv_directory)
for filename in glob.glob('{}/*'.format(tsv_directory)):
# avoid reading unexpected files, such as hidden files.
if not os.path.isfile(filename):
continue
logging.info(' - file: %s' % filename)
sentences = []
sid = 0
fields = {k: [] for k in cols.keys()}
with open(filename) as fin:
for line in fin.readlines():
if line.startswith('#'):
continue
l = line.split()
if l:
for k, v in fields.items():
v.append(l[cols[k]])
elif len(fields[TOK]) > 0:
wc += len(fields[TOK])
sentences.append(Sentence(fields))
fields = {k: [] for k in cols.keys()}
if len(fields[TOK]) > 0:
wc += len(fields[TOK])
sentences.append(Sentence(fields))
[[label_map.add(i) for i in sent[to_gold(key)]] for sent in sentences]
[sent.update({SID: i}) for i, sent in enumerate(sentences)]
sc += len(sentences)
documents.append(Document(sens=sentences))
[sent.update({DOC_ID: i}) for i, sent in enumerate(documents)]
logging.info('- dc = %d, sc = %d, wc = %d' % (len(documents), sc, wc))
return documents, label_map
def extract(self, did, sid, sen, **kwargs):
w = self.extract_sen(sen)
if self.label:
for i, label in enumerate(sen[to_gold(self.key)]):
self._data.append((self.extract_x(i,
w), self.extract_y(label)))
else:
for i in range(len(sen)):
self._data.append((self.extract_x(i, w), -1))
def extract_labels(self, sen):
if self.label:
return nd.array(
[self.label_map.cid(l) for l in sen[to_gold(self.key)]])
else:
return nd.array([-1 for _ in range(len(sen))])