def get_mnist_data_dict(unlabeled_samples, valid_set_size, test_set=False):
train_set = MNIST(("train",))
# Make sure the MNIST data is in right format
train_set.data_sources = (
(train_set.data_sources[0] / 255.).astype(numpy.float32),
train_set.data_sources[1])
# Take all indices and permutate them
all_ind = numpy.arange(train_set.num_examples)
rng = numpy.random.RandomState(seed=1)
rng.shuffle(all_ind)
data = AttributeDict()
# Choose the training set
data.train = train_set
data.train_ind = all_ind[:unlabeled_samples]
# Then choose validation set from the remaining indices
data.valid = train_set
data.valid_ind = numpy.setdiff1d(all_ind, data.train_ind)[:valid_set_size]
logger.info('Using %d examples for validation' % len(data.valid_ind))
# Only touch test data if requested
if test_set:
data.test = MNIST(("test",))
data.test_ind = numpy.arange(data.test.num_examples)
return data
