def test_get_batches():
data = {
'data0': [[i] * 2 for i in range(10)],
'data1': [[i] * 3 for i in range(10)]
}
batch_generator = batch.get_batches(data, batch_size=3, exact_epoch=True)
batches = list(batch_generator)
assert (batches[0]['data0'] == np.array([[6, 6], [3, 3], [0, 0]])).all()
assert (batches[0]['data1'] == np.array([[6, 6, 6], [3, 3, 3],
[0, 0, 0]])).all()
assert (batches[1]['data0'] == np.array([[4, 4], [5, 5], [2, 2]])).all()
assert (batches[1]['data1'] == np.array([[4, 4, 4], [5, 5, 5],
[2, 2, 2]])).all()
assert (batches[2]['data0'] == np.array([[1, 1], [9, 9], [8, 8]])).all()
assert (batches[2]['data1'] == np.array([[1, 1, 1], [9, 9, 9],
[8, 8, 8]])).all()
assert (batches[3]['data0'] == np.array([[7, 7]])).all()
assert (batches[3]['data1'] == np.array([[7, 7, 7]])).all()
assert len(batches) == 4
batch_generator = batch.get_batches(data, batch_size=3, exact_epoch=False)
batches = list(batch_generator)
assert len(batches) == 3
def dataset_generator(self, dataset: List[Tuple[QASetting, List[Answer]]],
is_eval: bool) -> Iterable[Mapping[TensorPort, np.ndarray]]:
corpus = self.preprocess(dataset)
xy_dict = {
Ports.Input.multiple_support: corpus["support"],
Ports.Input.question: corpus["question"],
Ports.Input.atomic_candidates: corpus["candidates"],
Ports.Targets.candidate_labels: corpus["targets"]
}
return get_batches(xy_dict)
def dataset_generator(
self,
dataset: List[Tuple[QASetting, List[Answer]]],
is_eval: bool,
test_time=False) -> Iterable[Mapping[TensorPort, np.ndarray]]:
corpus = self.preprocess(dataset, test_time=test_time)
xy_dict = {
Ports.Input.question: corpus["question"],
Ports.Input.atomic_candidates: corpus["candidates"],
Ports.Targets.target_index: corpus["answers"]
}
return get_batches(xy_dict, batch_size=self.config['batch_size'])
def dataset_generator(self, dataset: List[Tuple[QASetting, List[Answer]]],
is_eval: bool) -> Iterable[Mapping[TensorPort, np.ndarray]]:
corpus, _, _, _ = \
preprocess_with_pipeline(dataset,
self.shared_vocab_config.vocab, self.answer_vocab, use_single_support=True, sepvocab=True)
xy_dict = {
Ports.Input.single_support: corpus["support"],
Ports.Input.question: corpus["question"],
Ports.Targets.candidate_idx: corpus["answers"],
Ports.Input.question_length : corpus['question_lengths'],
Ports.Input.support_length : corpus['support_lengths'],
Ports.Input.sample_id : corpus['ids']
}
return get_batches(xy_dict)
if c == True:
for k, batch in enumerate(batches):
for kk, seq2 in enumerate(batch["relation_matrices"]):
for kkk, seq in enumerate(seq2):
for kkkk, tok in enumerate(seq):
batches[k]["relation_matrices"][kk][kkk][kkkk] = randint(0, len(rel_type_vocab) - 1)
return batches
if __name__ == "__main__":
reset_output_dir()
vocab = Vocab()
instances = read_ann(test_dir)
fill_vocab(instances, vocab)
batchable = convert_to_batchable_format(instances, vocab) #[:2]
#print(batchable)
batches = list(get_batches(batchable))#[:2]
# random baseline
batches = randomBaseline(batches, a=False, b=False, c=True)
for batch in batches:
convert_batch_to_ann(batch, instances, "/tmp")
remove_anno = "keys" # "", "types", "rel" or "keys"
calculateMeasures(test_dir, "/tmp/", remove_anno = remove_anno)
# print(instances[0].labels)