def setup_from_data(self, data: Iterable[Tuple[QASetting, List[Answer]]]):
vocab = self.shared_resources.vocab
if not vocab.frozen:
preprocessing.fill_vocab(
(q for q, _ in data),
vocab,
lowercase=self.shared_resources.config.get('lowercase', True))
vocab.freeze()
if vocab.emb is not None:
self.shared_resources.embeddings = np.zeros(
[len(vocab), vocab.emb_length])
for w, i in self.shared_resources.vocab.sym2id.items():
e = vocab.emb.get(w)
if e is not None:
self.shared_resources.embeddings[i] = e
if not hasattr(self.shared_resources, 'answer_vocab'
) or not self.shared_resources.answer_vocab.frozen:
self.shared_resources.answer_vocab = util.create_answer_vocab(
qa_settings=(q for q, _ in data),
answers=(a for _, ass in data for a in ass))
self.shared_resources.answer_vocab.freeze()
self.shared_resources.config['answer_size'] = len(
self.shared_resources.answer_vocab)
self.shared_resources.char_vocab = preprocessing.char_vocab_from_vocab(
self.shared_resources.vocab)
def setup_from_data(self, data: Iterable[Tuple[QASetting, List[Answer]]]):
# create character vocab + word lengths + char ids per word
if not self.shared_resources.vocab.frozen:
preprocessing.fill_vocab(
(q for q, _ in data), self.shared_resources.vocab,
self.shared_resources.config.get("lowercase", False))
self.shared_resources.vocab.freeze()
self.shared_resources.char_vocab = preprocessing.char_vocab_from_vocab(
self.shared_resources.vocab)
def setup_from_data(self, data: Iterable[Tuple[QASetting, List[Answer]]]):
vocab = self.shared_resources.vocab
if not vocab.frozen:
preprocessing.fill_vocab(
(q for q, _ in data), vocab, lowercase=self.shared_resources.config.get('lowercase', True))
vocab.freeze()
if not hasattr(self.shared_resources, 'answer_vocab') or not self.shared_resources.answer_vocab.frozen:
self.shared_resources.answer_vocab = util.create_answer_vocab(
qa_settings=(q for q, _ in data), answers=(a for _, ass in data for a in ass))
self.shared_resources.answer_vocab.freeze()
self.shared_resources.char_vocab = preprocessing.char_vocab_from_vocab(self.shared_resources.vocab)
def setup_from_data(self, data: Iterable[Tuple[QASetting, List[Answer]]]):
if not self.shared_resources.vocab.frozen:
self.shared_resources.vocab = preprocessing.fill_vocab(
(q for q, _ in data), self.shared_resources.vocab, lowercase=True)
self.shared_resources.vocab.freeze()
if not hasattr(self.shared_resources, 'answer_vocab') or not self.shared_resources.answer_vocab.frozen:
self.shared_resources.answer_vocab = create_answer_vocab(answers=(a for _, ass in data for a in ass))
self.shared_resources.answer_vocab.freeze()
self.shared_resources.config['answer_size'] = self.shared_resources.config.get(
'answer_size', len(self.shared_resources.answer_vocab))
self.shared_resources.char_vocab = {chr(i): i for i in range(256)}
def test_vocab():
train_data = [
QASetting(question='A person is training his horse for a competition.',
support=['A person on a horse jumps over a broken down airplane.'],
candidates=['entailment', 'neutral', 'contradiction'])
]
print('build vocab based on train data')
train_vocab = preprocessing.fill_vocab(train_data)
train_vocab.freeze()
pprint(train_vocab._sym2freqs)
pprint(train_vocab._sym2id)
MIN_VOCAB_FREQ, MAX_VOCAB_CNT = 2, 10
train_vocab = train_vocab.prune(MIN_VOCAB_FREQ, MAX_VOCAB_CNT)
pprint(train_vocab._sym2freqs)
pprint(train_vocab._sym2id)
print('encode train data')
train_data = preprocessing.nlp_preprocess(train_data[0].question, train_vocab)[0]
print(train_data)