def _fill_names_batches(self, shuffle):
'''Create the desired batches of sequences
* Select the desired sequences according to seq_per_subset
* Set self.nsamples, self.nbatches and self.names_batches.
* Set self.names_batches, an iterator over the batches of names.
'''
# self.names_sequences has the following structure:
# key: a subset
# value: a tuple of *sequences* with structure
# `((subset, filename1), (subset, filename2), ..)`
# and length `seq_length`.
# E.g.,
# names_sequences['0001TP'][0] = (('0001TP', '0001TP_006690.png'),)
#
# Here we copy it in `names_sequences`, selecting a random subset
# of values per key in case a limit has been imposed via
# self.seq_per_subset.
names_sequences = OrderedDict()
for prefix, sequences in self.names_sequences.items():
# Pick only a subset of sequences per each video
if self.seq_per_subset:
# Pick `seq_per_subset` random indices
idx = np.random.permutation(range(
len(sequences)))[:self.seq_per_subset]
# Select only those sequences
sequences = np.array(sequences)[idx]
names_sequences.setdefault(prefix, []).extend(sequences)
# Group the sequences into minibatches of `batch_size` length
if self.one_subset_per_batch:
# Group each subset separately
names_batches = [
el for seq in names_sequences.itervalues()
for el in grouper(seq, self.batch_size)
]
if shuffle:
self.rng.shuffle(names_batches) # shuffle the batches
else:
# Concat the values in a unique list, get rid of the keys
names_sequences_flat = [
el for outer_el in names_sequences.values() for el in outer_el
]
if shuffle:
self.rng.shuffle(names_sequences_flat) # shuffle the sequences
# Group all the subsets together
names_batches = [
el for el in grouper(names_sequences_flat, self.batch_size)
]
self.nsamples = len(names_sequences)
self.nbatches = len(names_batches)
# `names_batches` contains three nested tuples and has shape
# (batch_size, seq_length, 2), where the most inner element is a
# tuple `(subset, filename)`.
self.names_batches = iter(names_batches)
def _fill_names_batches(self, shuffle):
'''Create the desired batches of sequences
* Select the desired sequences according to the parameters
* Set self.nsamples, self.nbatches and self.names_batches.
* Set self.names_batches, an iterator over the batches of names.
'''
names_sequences = []
for prefix, sequences in self.names_sequences.items():
# Pick only a subset of sequences per each video
if self.seq_per_subset:
# Pick `seq_per_subset` random indices
idx = np.random.permutation(range(len(sequences)))[
:self.seq_per_subset]
# Select only those sequences
sequences = np.array(sequences)[idx]
names_sequences.extend(sequences)
# Shuffle the sequences
if shuffle:
self.rng.shuffle(names_sequences)
# Group the sequences into minibatches
names_batches = [el for el in grouper(names_sequences,
self.batch_size)]
self.nsamples = len(names_sequences)
self.nbatches = len(names_batches)
self.names_batches = iter(names_batches)
def _fill_names_batches(self, shuffle):
'''Create the desired batches of sequences
* Select the desired sequences according to seq_per_subset
* Set self.nsamples, self.nbatches and self.names_batches.
* Set self.names_batches, an iterator over the batches of names.
'''
names_sequences = OrderedDict()
for prefix, sequences in self.names_sequences.items():
# Pick only a subset of sequences per each video
if self.seq_per_subset:
# Pick `seq_per_subset` random indices
idx = np.random.permutation(range(
len(sequences)))[:self.seq_per_subset]
# Select only those sequences
sequences = np.array(sequences)[idx]
names_sequences.setdefault(prefix, []).extend(sequences)
# Group the sequences into minibatches
if self.one_subset_per_batch:
names_batches = [
el for seq in names_sequences.itervalues()
for el in grouper(seq, self.batch_size)
]
if shuffle:
self.rng.shuffle(names_batches) # shuffle the batches
else:
names_sequences = [
el for outer_el in names_sequences.values() for el in outer_el
]
if shuffle:
self.rng.shuffle(names_sequences) # shuffle the sequences
names_batches = [
el for el in grouper(names_sequences, self.batch_size)
]
self.nsamples = len(names_sequences)
self.nbatches = len(names_batches)
self.names_batches = iter(names_batches)
def _fill_names_batches(self, shuffle):
'''Create the desired batches of sequences
* Select the desired sequences according to seq_per_subset
* Set self.nsamples, self.nbatches and self.names_batches.
* Set self.names_batches, an iterator over the batches of names.
'''
names_sequences = OrderedDict()
for prefix, sequences in self.names_sequences.items():
# Pick only a subset of sequences per each video
if self.seq_per_subset:
# Pick `seq_per_subset` random indices
idx = np.random.permutation(range(len(sequences)))[
:self.seq_per_subset]
# Select only those sequences
sequences = np.array(sequences)[idx]
names_sequences.setdefault(prefix, []).extend(sequences)
# Group the sequences into minibatches
if self.one_subset_per_batch:
names_batches = [el for seq in names_sequences.itervalues()
for el in grouper(seq, self.batch_size)]
if shuffle:
self.rng.shuffle(names_batches) # shuffle the batches
else:
names_sequences = [el for outer_el in names_sequences.values()
for el in outer_el]
if shuffle:
self.rng.shuffle(names_sequences) # shuffle the sequences
names_batches = [el for el in grouper(names_sequences,
self.batch_size)]
self.nsamples = len(names_sequences)
self.nbatches = len(names_batches)
self.names_batches = iter(names_batches)