def test_caption_random_features(train_params: str, translate_params: str):
# generate random names
source_list = [
''.join(random.choice(string.ascii_uppercase) for _ in range(4))
for i in range(15)
]
prefix = "tmp_caption_random"
use_features = True
with tmp_img_captioning_dataset(source_list,
prefix,
train_max_length=_LINE_MAX_LENGTH,
dev_max_length=_LINE_MAX_LENGTH,
test_max_length=_TEST_MAX_LENGTH,
use_features=use_features) as data:
# Test model configuration, including the output equivalence of batch and no-batch decoding
translate_params_batch = translate_params + " --batch-size 2"
# Ignore return values (perplexity and BLEU) for integration test
run_train_captioning(train_params=train_params,
translate_params=translate_params,
translate_params_equiv=translate_params_batch,
train_source_path=data['source'],
train_target_path=data['target'],
dev_source_path=data['validation_source'],
dev_target_path=data['validation_target'],
test_source_path=data['test_source'],
test_target_path=data['test_target'],
max_seq_len=_LINE_MAX_LENGTH + 1,
work_dir=data['work_dir'])
def test_caption_random_features(train_params: str, translate_params: str):
# generate random names
source_list = [''.join(random.choice(string.ascii_uppercase) for _ in range(4)) for i in range(15)]
prefix = "tmp_caption_ramdom"
use_features = True
with tmp_img_captioning_dataset(source_list,
prefix,
train_max_length=_LINE_MAX_LENGTH,
dev_max_length=_LINE_MAX_LENGTH,
test_max_length=_TEST_MAX_LENGTH,
use_features=use_features) as data:
# Test model configuration, including the output equivalence of batch and no-batch decoding
translate_params_batch = translate_params + " --batch-size 2"
# Ignore return values (perplexity and BLEU) for integration test
run_train_captioning(train_params=train_params,
translate_params=translate_params,
translate_params_equiv=translate_params_batch,
train_source_path=data['source'],
train_target_path=data['target'],
dev_source_path=data['validation_source'],
dev_target_path=data['validation_target'],
test_source_path=data['test_source'],
test_target_path=data['test_target'],
max_seq_len=_LINE_MAX_LENGTH + 1,
work_dir=data['work_dir'])
def test_caption_random_features(layer: str):
source_image_size = (3, 20, 20)
batch_size = 8
extract_params = "--source-image-size {s1} {s2} {s3} --batch-size {batch_size} " \
"--image-encoder-layer {layer}".format(s1=source_image_size[0],
s2=source_image_size[1],
s3=source_image_size[2],
batch_size=batch_size,
layer=layer)
# generate random names
source_list = [
''.join(random.choice(string.ascii_uppercase) for _ in range(4)) for i
in range(8)]
prefix = "tmp_features"
use_features = False
with tmp_img_captioning_dataset(source_list,
prefix,
train_max_length=1,
dev_max_length=1,
test_max_length=1,
use_features=use_features) as data:
source_files = [data["source"], data["validation_source"],
data["test_source"]]
run_extract_features_captioning(source_image_size=source_image_size,
batch_size=batch_size,
extract_params=extract_params,
source_files=source_files,
image_root=data['work_dir'])
def test_caption_random_features(layer: str):
source_image_size = (3, 20, 20)
batch_size = 8
extract_params = "--source-image-size {s1} {s2} {s3} --batch-size {batch_size} " \
"--image-encoder-layer {layer}".format(s1=source_image_size[0],
s2=source_image_size[1],
s3=source_image_size[2],
batch_size=batch_size,
layer=layer)
# generate random names
source_list = [
''.join(random.choice(string.ascii_uppercase) for _ in range(4))
for i in range(8)
]
prefix = "tmp_features"
use_features = False
with tmp_img_captioning_dataset(source_list,
prefix,
train_max_length=1,
dev_max_length=1,
test_max_length=1,
use_features=use_features) as data:
source_files = [
data["source"], data["validation_source"], data["test_source"]
]
run_extract_features_captioning(source_image_size=source_image_size,
batch_size=batch_size,
extract_params=extract_params,
source_files=source_files,
image_root=data['work_dir'])
def test_get_training_image_text_data_iters():
# Test images
source_list = ['1', '2', '3', '4', '100']
prefix = "tmp_corpus"
use_feature_loader = False
preload_features = False
train_max_length = 30
dev_max_length = 30
expected_mean = 1.0
expected_std = 1.0
test_max_length = 30
batch_size = 5
if use_feature_loader:
source_image_size = _FEATURE_SHAPE
else:
source_image_size = _CNN_INPUT_IMAGE_SHAPE
with tmp_img_captioning_dataset(source_list, prefix, train_max_length,
dev_max_length, test_max_length,
use_feature_loader) as data:
# tmp common vocab
vcb = vocab.build_from_paths([data['target'], data['target']])
train_iter, val_iter, config_data, data_info = data_io.get_training_image_text_data_iters(
source_root=data['work_dir'],
source=data['source'],
target=data['target'],
validation_source_root=data['work_dir'],
validation_source=data['validation_source'],
validation_target=data['validation_target'],
vocab_target=vcb,
vocab_target_path=None,
batch_size=batch_size,
batch_by_words=False,
batch_num_devices=1,
source_image_size=source_image_size,
fill_up="replicate",
max_seq_len_target=train_max_length,
bucketing=False,
bucket_width=10,
use_feature_loader=use_feature_loader,
preload_features=preload_features)
assert isinstance(train_iter, data_io.ParallelSampleIter)
assert isinstance(val_iter, data_io.ParallelSampleIter)
assert isinstance(config_data, data_io.DataConfig)
assert isinstance(data_info.sources[0], data_io.FileListReader)
assert data_info.target == data['target']
assert data_info.source_vocabs is None
assert data_info.target_vocab is None
assert config_data.data_statistics.max_observed_len_source == 0
assert config_data.data_statistics.max_observed_len_target == train_max_length - 1
assert np.isclose(config_data.data_statistics.length_ratio_mean,
expected_mean)
assert np.isclose(config_data.data_statistics.length_ratio_std,
expected_std)
assert train_iter.batch_size == batch_size
assert val_iter.batch_size == batch_size
assert train_iter.default_bucket_key == (0, train_max_length)
assert val_iter.default_bucket_key == (0, dev_max_length)
assert train_iter.dtype == 'float32'
# test some batches
bos_id = vcb[C.BOS_SYMBOL]
expected_first_target_symbols = np.full((batch_size, ),
bos_id,
dtype='float32')
for epoch in range(2):
while train_iter.iter_next():
batch = train_iter.next()
assert len(batch.data) == 2
assert len(batch.label) == 1
assert batch.bucket_key in train_iter.buckets
source = batch.data[0].asnumpy()
target = batch.data[1].asnumpy()
label = batch.label[0].asnumpy()
assert source.shape[0] == target.shape[0] == label.shape[
0] == batch_size
# target first symbol should be BOS
assert np.array_equal(target[:, 0],
expected_first_target_symbols)
# label first symbol should be 2nd target symbol
assert np.array_equal(label[:, 0], target[:, 1])
# each label sequence contains one EOS symbol
assert np.sum(label == vcb[C.EOS_SYMBOL]) == batch_size
train_iter.reset()
def test_get_training_image_text_data_iters():
# Test images
source_list = ['1', '2', '3', '4', '100']
prefix = "tmp_corpus"
use_feature_loader = False
preload_features = False
train_max_length = 30
dev_max_length = 30
expected_mean = 1.0
expected_std = 1.0
test_max_length = 30
batch_size = 5
if use_feature_loader:
source_image_size = _FEATURE_SHAPE
else:
source_image_size = _CNN_INPUT_IMAGE_SHAPE
with tmp_img_captioning_dataset(source_list,
prefix,
train_max_length,
dev_max_length,
test_max_length,
use_feature_loader) as data:
# tmp common vocab
vcb = vocab.build_from_paths([data['target'], data['target']])
train_iter, val_iter, config_data, data_info = data_io.get_training_image_text_data_iters(source_root=data['work_dir'],
source=data['source'],
target=data['target'],
validation_source_root=data['work_dir'],
validation_source=data['validation_source'],
validation_target=data['validation_target'],
vocab_target=vcb,
vocab_target_path=None,
batch_size=batch_size,
batch_by_words=False,
batch_num_devices=1,
source_image_size=source_image_size,
fill_up="replicate",
max_seq_len_target=train_max_length,
bucketing=False,
bucket_width=10,
use_feature_loader=use_feature_loader,
preload_features=preload_features)
assert isinstance(train_iter, data_io.ParallelSampleIter)
assert isinstance(val_iter, data_io.ParallelSampleIter)
assert isinstance(config_data, data_io.DataConfig)
assert isinstance(data_info.sources[0], data_io.FileListReader)
assert data_info.target == data['target']
assert data_info.source_vocabs is None
assert data_info.target_vocab is None
assert config_data.data_statistics.max_observed_len_source == 0
assert config_data.data_statistics.max_observed_len_target == train_max_length - 1
assert np.isclose(config_data.data_statistics.length_ratio_mean, expected_mean)
assert np.isclose(config_data.data_statistics.length_ratio_std, expected_std)
assert train_iter.batch_size == batch_size
assert val_iter.batch_size == batch_size
assert train_iter.default_bucket_key == (0, train_max_length)
assert val_iter.default_bucket_key == (0, dev_max_length)
assert train_iter.dtype == 'float32'
# test some batches
bos_id = vcb[C.BOS_SYMBOL]
expected_first_target_symbols = np.full((batch_size,), bos_id, dtype='float32')
for epoch in range(2):
while train_iter.iter_next():
batch = train_iter.next()
assert len(batch.data) == 2
assert len(batch.label) == 1
assert batch.bucket_key in train_iter.buckets
source = batch.data[0].asnumpy()
target = batch.data[1].asnumpy()
label = batch.label[0].asnumpy()
assert source.shape[0] == target.shape[0] == label.shape[0] == batch_size
# target first symbol should be BOS
assert np.array_equal(target[:, 0], expected_first_target_symbols)
# label first symbol should be 2nd target symbol
assert np.array_equal(label[:, 0], target[:, 1])
# each label sequence contains one EOS symbol
assert np.sum(label == vcb[C.EOS_SYMBOL]) == batch_size
train_iter.reset()