def prepare_training_data(src_files, src_files_encoding, trg_files,
trg_files_encoding, src_output_file,
trg_output_file):
'''
for each pair of source/target files, check they have the same number of sentences;
do shuffle and save with utf-8 encodings
'''
src = chain(
*[iter_(open_(f, encoding=src_files_encoding)) for f in src_files])
trg = chain(
*[iter_(open_(f, encoding=trg_files_encoding)) for f in trg_files])
# TODO: find a way not to load all sentences into memory
logger.info("reading sentences from source files...")
src_sentences = [sent for sent in src]
logger.info("reading sentences from target files...")
trg_sentences = [sent for sent in trg]
assert len(src_sentences) == len(trg_sentences)
logger.info("number of sentences:%d" % len(src_sentences))
# '\n' has been removed from a sentence
assert src_sentences[0].endswith('\n')
# do the shuffle
ids = list(range(len(src_sentences)))
random.shuffle(ids)
with codecs.open(src_output_file, 'w', 'UTF-8') as f_src:
with codecs.open(trg_output_file, 'w', 'UTF-8') as f_trg:
for i in ids:
f_src.write(src_sentences[i])
f_trg.write(trg_sentences[i])
def test_dict_iterator():
d = {'a': 'b', 1: 2}
assert list(iter_(d)) == list(iter(d))
assert list(iter_(d.items())) == list(iter(d.items()))
assert list(iter_(d.keys())) == list(iter(d.keys()))
assert list(iter_(d.values())) == list(iter(d.values()))
yield verify_pickle, iter_, iter, 2, 1, d
yield verify_pickle, iter_, iter, 2, 1, d.items()
yield verify_pickle, iter_, iter, 2, 1, d.values()
yield verify_pickle, iter_, iter, 2, 1, d.keys()
def test_dict_iterator():
d = {'a': 'b', 1: 2}
assert list(iter_(d)) == list(iter(d))
assert list(iter_(d.items())) == list(iter(d.items()))
assert list(iter_(d.keys())) == list(iter(d.keys()))
assert list(iter_(d.values())) == list(iter(d.values()))
yield verify_pickle, iter_, iter, 2, 1, d
yield verify_pickle, iter_, iter, 2, 1, d.items()
yield verify_pickle, iter_, iter, 2, 1, d.values()
yield verify_pickle, iter_, iter, 2, 1, d.keys()
def get_request_iterator(self):
l = [
f for f in os.listdir(self.path)
if os.path.isfile(os.path.join(self.path, f))
]
if self.shuffle:
random.shuffle(l)
return iter_(l)
def get_request_iterator(self):
i = range(self.nitems)
j = range(self.nfeats)
if self.irandom:
random.shuffle(i)
if self.jrandom:
random.shuffle(j)
ib = batch(i, self.ibatchsize)
jb = batch(j, self.jbatchsize)
return iter_([(ii, jj) for ii in ib for jj in jb])
def get_request_iterator(self):
i = range(self.nitems)
j = range(self.nfeats)
if self.irandom:
random.shuffle(i)
if self.jrandom:
random.shuffle(j)
ib = batch(i, self.ibatchsize)
jb = batch(j, self.jbatchsize)
return iter_([(ii, jj) for ii in ib for jj in jb])
def get_request_iterator(self):
self.refs = list(self.squad_iterator.get_request_iterator())
cnn_refs = list(self.cnn_iterator.get_request_iterator())
random.shuffle(cnn_refs)
cnn_to_add = int(self.cnn_ratio * len(self.refs))
if cnn_to_add > len(cnn_refs):
print "To many CNN data points requested"
cnn_to_add = len(cnn_refs)
self.refs += cnn_refs[:cnn_to_add]
if self.shuffle:
print "Shuffling CNN and SQuAD (should occur every epoch)"
random.shuffle(self.refs)
return iter_(self.refs)
def get_request_iterator(self):
cuts = self.use_cuts
if cuts == None:
cuts = [random.randrange(first_time, last_time) for _ in range(self.num_cuts)]
l = []
with sqlite3.connect(self.dbfile) as db:
c = db.cursor()
for cut in cuts:
part = [i for (i,) in
c.execute('SELECT trip FROM trip_times WHERE begin >= ? AND begin <= ? AND end >= ?',
(cut - 40000, cut, cut))]
l = l + part
random.shuffle(l)
return iter_(l)
def get_request_iterator(self):
cuts = self.use_cuts
if cuts == None:
cuts = [
random.randrange(first_time, last_time)
for _ in range(self.num_cuts)
]
l = []
with sqlite3.connect(self.dbfile) as db:
c = db.cursor()
for cut in cuts:
part = [
i for (i, ) in c.execute(
'SELECT trip FROM trip_times WHERE begin >= ? AND begin <= ? AND end >= ?',
(cut - 40000, cut, cut))
]
l = l + part
random.shuffle(l)
return iter_(l)
def open(self): return chain(*[iter_( codecs.open(f, encoding="latin1") ) for f in self.files])
def get_request_iterator(self):
return iter_(self.indices)
def open(self):
return chain(*[iter_(open_(f, encoding=self.encoding))
for f in self.files])
def open(self):
return chain(*[iter_(open(f)) for f in self.files])
def test_file_iterator_pickling():
f = _create_test_file()
it = iter_(open(f.name))
last = [next(it) for _ in range(2)][-1]
first = next(cPickle.loads(cPickle.dumps(it)))
assert int(first) == int(last) + 1
def open(self):
return chain(
*[iter_(codecs.open(f, encoding="latin1")) for f in self.files])
def tee(iterable, n=2):
"""tee(iterable, n=2) --> tuple of n independent iterators."""
return tee_manager(iter_(iterable), n=n).iterators()
def open(self):
return chain(*[iter_(open(f)) for f in self.files])
def open(self):
return bAbIState(chain(*[iter_(open(f)) for f in self.files]), [], 0)
def __init__(self, iterable, n=2):
self._iterable = iter_(iterable)
self._deques = tuple(collections.deque() for _ in range(n))
def get_request_iterator(self):
return iter_(xrange(self.num_examples))
def get_request_iterator(self):
l = [(random.randrange(0, self.item_range - self.seq_len + 1), self.seq_len)
for _ in xrange(self.num_seqs)]
return iter_(l)
def open(self):
handlers = [open(f, "rb") for f in self.files]
return chain(*[iter_(h) for h in handlers]), handlers
def get_request_iterator(self):
indices = list(self.indices)
self.rng.shuffle(indices)
return iter_(indices)
def get_request_iterator(self):
indices = list(range(self.num_examples))
self.rng.shuffle(indices)
return imap(list, imap(
islice, repeat(iter_(indices), self.num_batches),
repeat(self.batch_size, self.num_batches)))
def get_request_iterator(self):
if self.shuffle:
random.shuffle(self.reference)
return iter_(self.reference)
def open(self):
iterators = [iter_(channel) for channel in self.iterables]
return izip(*iterators)
def get_request_iterator(self):
return iter_(list(range(2)))
def get_request_iterator(self):
l = [f for f in os.listdir(self.path)
if os.path.isfile(os.path.join(self.path, f))]
if self.shuffle:
random.shuffle(l)
return iter_(l)
def __init__(self, n_list, seq):
self._n_list = iter_(n_list)
self._seq = iter_(seq)
def open(self):
return iter_(self.piano_rolls)
def open(self):
iterators = [iter_(channel) for channel in self.iterables]
return izip(*iterators)
def get_request_iterator(self):
indices = list(range(self.num_examples))
self.rng.shuffle(indices)
return iter_(indices)
def get_request_iterator(self):
it = self.iteration_scheme.get_request_iterator()
return iter_([it.next() for n in range(self.times)])
def test_file_iterator_pickling():
f = _create_test_file()
it = iter_(open(f.name))
last = [next(it) for _ in range(2)][-1]
first = next(cPickle.loads(cPickle.dumps(it)))
assert int(first) == int(last) + 1
def open(self):
return chain(
*[iter_(open_(f, encoding=self.encoding)) for f in self.files])
def open(self):
return iter_(self.files)
