def prepare_glove(self, dimension):
if self.index is not None: return
if not os.path.exists(join(get_data_path(), 'glove', 'index_50.p')):
dims = [50, 100, 200, 300]
base_filename = 'glove.6B.{0}d.txt'
paths = [
join(get_data_path(), 'glove', base_filename.format(dim))
for dim in dims
]
for path, dim in zip(paths, dims):
index = {}
index = {'PATH': path}
with open(path, 'rb') as f:
log.info('Building index for {0}', path)
while True:
prev_pos = f.tell()
line = f.readline().decode('utf-8')
if line == '': break
next_pos = f.tell()
data = line.strip().split(' ')
token = data[0]
index[token] = (prev_pos, next_pos)
log.info('Saving glove index...')
json.dump(
index,
open(
join(get_data_path(), 'glove',
'index_{0}.p'.format(dim)), 'w'))
log.info('Loading glove index...')
self.index = json.load(
open(
join(get_data_path(), 'glove',
'index_{0}.p'.format(dimension)), 'r'))
def download_glove(self):
if not os.path.exists(join(get_data_path(), 'glove')):
log.info('Glove data is missing, dowloading data now...')
os.mkdir(join(get_data_path(), 'glove'))
bashmagic.wget("http://nlp.stanford.edu/data/glove.6B.zip",
join(get_data_path(), 'glove'))
bashmagic.unzip(join(get_data_path(), 'glove', 'glove.6B.zip'),
join(get_data_path(), 'glove'))
def test_save_load_data(dtype):
folder = join(get_data_path(), 'test_hdf')
if os.path.exists(folder):
shutil.rmtree(folder)
os.mkdir(folder)
for i in range(5):
filename = str(uuid.uuid4())
data1 = dtype(np.random.randn(100, 100))
save_data(join(folder, filename), data1)
data2 = load_data(join(folder, filename))
np.testing.assert_array_equal(data1, data2, 'Arrays must be equal')
shutil.rmtree(folder)
def preprocess_SNLI(delete_data=False):
# load data
#names, file_paths = snli2json()
#train_path, dev_path, test_path = file_paths
tokenizer = nltk.tokenize.WordPunctTokenizer()
zip_path = join(get_data_path(), 'snli_1.0.zip', 'snli_1.0')
file_paths = [
'snli_1.0_train.jsonl', 'snli_1.0_dev.jsonl', 'snli_1.0_test.jsonl'
]
not_t = []
t = ['input', 'support', 'target']
# tokenize and convert to hdf5
# 1. Setup pipeline to save lengths and generate vocabulary
p = Pipeline('snli_example', delete_data)
p.add_path(join(zip_path, file_paths[0]))
p.add_line_processor(JsonLoaderProcessors())
p.add_line_processor(
RemoveLineOnJsonValueCondition('gold_label',
lambda label: label == '-'))
p.add_line_processor(
DictKey2ListMapper(['sentence1', 'sentence2', 'gold_label']))
p.add_sent_processor(ToLower())
p.add_sent_processor(Tokenizer(tokenizer.tokenize), t)
#p.add_sent_processor(NaiveNCharTokenizer(3), not_t)
p.add_token_processor(AddToVocab())
p.add_post_processor(SaveLengthsToState())
p.execute()
p.clear_processors()
p.state['vocab'].save_to_disk()
# 2. Process the data further to stream it to hdf5
p.add_sent_processor(ToLower())
p.add_sent_processor(Tokenizer(tokenizer.tokenize), t)
#p.add_sent_processor(NaiveNCharTokenizer(3), not_t)
p.add_post_processor(ConvertTokenToIdx())
p.add_post_processor(
CreateBinsByNestedLength('snli_train', min_batch_size=128))
state = p.execute()
# dev and test data
p2 = Pipeline('snli_example')
p2.copy_vocab_from_pipeline(p)
p2.add_path(join(zip_path, file_paths[1]))
p2.add_line_processor(JsonLoaderProcessors())
p2.add_line_processor(
RemoveLineOnJsonValueCondition('gold_label',
lambda label: label == '-'))
p2.add_line_processor(
DictKey2ListMapper(['sentence1', 'sentence2', 'gold_label']))
p2.add_sent_processor(ToLower())
p2.add_sent_processor(Tokenizer(tokenizer.tokenize), t)
#p2.add_sent_processor(NaiveNCharTokenizer(3), not_t)
p2.add_post_processor(SaveLengthsToState())
p2.execute()
p2.clear_processors()
p2.add_sent_processor(ToLower())
p2.add_sent_processor(Tokenizer(tokenizer.tokenize), t)
#p2.add_sent_processor(NaiveNCharTokenizer(3), not_t)
p2.add_post_processor(ConvertTokenToIdx())
p2.add_post_processor(StreamToHDF5('snli_dev'))
p2.execute()
p3 = Pipeline('snli_example')
p3.copy_vocab_from_pipeline(p)
p3.add_path(join(zip_path, file_paths[2]))
p3.add_line_processor(JsonLoaderProcessors())
p3.add_line_processor(
RemoveLineOnJsonValueCondition('gold_label',
lambda label: label == '-'))
p3.add_line_processor(
DictKey2ListMapper(['sentence1', 'sentence2', 'gold_label']))
p3.add_sent_processor(ToLower())
p3.add_sent_processor(Tokenizer(tokenizer.tokenize), t)
#p3.add_sent_processor(NaiveNCharTokenizer(3), not_t)
p3.add_post_processor(SaveLengthsToState())
p3.execute()
p3.clear_processors()
p3.add_sent_processor(ToLower())
p3.add_sent_processor(Tokenizer(tokenizer.tokenize), t)
#p3.add_sent_processor(NaiveNCharTokenizer(3), not_t)
p3.add_post_processor(ConvertTokenToIdx())
p3.add_post_processor(StreamToHDF5('snli_test'))
p3.execute()
def __init__(self,
pipeline_name,
name,
batch_size,
mnt_name='',
loader_threads=4,
randomize=False,
seed=None,
keys=['input', 'support', 'target'],
is_volatile=False,
cache_size_GB=4):
config_path = join(get_data_path(mnt_name=mnt_name), pipeline_name,
name, 'hdf5_config.pkl')
if not exists(config_path):
log.error(
'Path {0} does not exists! Have you forgotten to preprocess your dataset?',
config_path)
config = pickle.load(open(config_path, 'rb'))
self.paths = config['paths']
self.fractions = config['fractions']
self.num_batches = int(np.sum(config['counts']) / batch_size)
self.max_lengths = config['max_lengths']
self.batch_size = batch_size
self.batch_idx = 0
self.prefetch_batch_idx = 0
self.loaders = []
self.prepared_batches = {}
self.prepared_batchidx = queue.Queue()
self.work = queue.Queue()
self.cached_batches = {}
self.end_iter_observers = []
self.end_epoch_observers = []
self.start_epoch_observers = []
self.at_batch_prepared_observers = []
self.state = BatcherState()
self.current_iter = 0
self.current_epoch = 0
self.timer = Timer()
self.loader_threads = loader_threads
if Config.backend == Backends.TORCH:
from spodernet.backends.torchbackend import TorchConverter, TorchCUDAConverter
self.subscribe_to_batch_prepared_event(DictConverter(keys))
self.subscribe_to_batch_prepared_event(TorchConverter(is_volatile))
if Config.cuda:
import torch
self.subscribe_to_batch_prepared_event(
TorchCUDAConverter(torch.cuda.current_device()))
elif Config.backend == Backends.TENSORFLOW:
from spodernet.backends.tfbackend import TensorFlowConverter
self.subscribe_to_batch_prepared_event(TensorFlowConverter())
elif Config.backend == Backends.TEST:
pass
elif Config.backend == Backends.CNTK:
self.subscribe_to_batch_prepared_event(DictConverter(keys))
else:
raise Exception('Backend has unsupported value {0}'.format(
Config.backend))
batchidx2paths, batchidx2start_end, shard2batchidx = self.create_batchidx_maps(
config['counts'])
for i in range(loader_threads):
seed = 2345 + (i * 83)
self.loaders.append(
DataLoaderSlave(self, batchidx2paths, batchidx2start_end,
randomize, self.paths, shard2batchidx, seed,
self.fractions, cache_size_GB))
self.loaders[-1].start()
