def dataset_summary(data_loader):
"""Create a histogram of class membership distribution within a dataset.
It is important to examine our training, validation, and test
datasets, to make sure that they are balanced.
"""
msglogger.info("Analyzing dataset:")
print_frequency = 50
for batch, (input, label_batch) in enumerate(data_loader):
try:
all_labels = np.append(all_labels, distiller.to_np(label_batch))
except NameError:
all_labels = distiller.to_np(label_batch)
if (batch + 1) % print_frequency == 0:
# progress indicator
print("batch: %d" % batch)
hist = np.histogram(all_labels, bins=np.arange(1000 + 1))
nclasses = len(hist[0])
for data_class, size in enumerate(hist[0]):
msglogger.info("\tClass {} = {}".format(data_class, size))
msglogger.info("Dataset contains {} items".format(len(
data_loader.sampler)))
msglogger.info("Found {} classes".format(nclasses))
msglogger.info("Average: {} samples per class".format(np.mean(hist[0])))
def test_threshold_mask():
# Create a 4-D tensor of 1s
a = torch.ones(3, 64, 32, 32)
# Change one element
a[1, 4, 17, 31] = 0.2
# Create and apply a mask
mask = distiller.threshold_mask(a, threshold=0.3)
assert np.sum(distiller.to_np(mask)) == (distiller.volume(a) - 1)
assert mask[1, 4, 17, 31] == 0
assert common.almost_equal(distiller.sparsity(mask), 1/distiller.volume(a))