def test_bernoulli_csl():
print 'loading MNIST test set'
# (train_x, _,
# valid_x, _,
# test_x, _) = data_provider.load_mnist(binary=False, standard_split=False)
# (train_x_b, _,
# valid_x_b, _,
# test_x_b, _) = data_provider.load_mnist(binary=True, standard_split=False)
(_, _), (_, _), (test_x, _) = data_tools.load_mnist('../../data')
(_, _), (_, _), (test_x_b, _) = data_tools.load_mnist_binary('../../data')
means = test_x.astype('float32')
means = numpy.clip(means, 1e-10, (1 - (1e-5)))
#means = numpy.random.uniform(size=(10000,784)).astype('float32') * 0 + 0.5
csl = CSL()
minibatches = test_x_b.reshape((1000, 10, 784)).astype('float32')
if 0:
# when means is a matrix of (N,D), representing only 1 chain
csl_fn = csl.get_CSL_fn_independent_Bernoulli_v2(means)
csl.compute_CSL_with_minibatches_one_chain(csl_fn, minibatches)
else:
# when means is a 3D tensor (N, K, D)
# When there are N chains, each chain having K samples of dimension D
chains = means.reshape(10, 100, 10, 784)
csl_fn = csl.get_CSL_fn_independent_Bernoulli()
csl.compute_CSL_with_minibatches(csl_fn, minibatches, chains)
def test_bernoulli_csl():
print 'loading MNIST test set'
# (train_x, _,
# valid_x, _,
# test_x, _) = data_provider.load_mnist(binary=False, standard_split=False)
# (train_x_b, _,
# valid_x_b, _,
# test_x_b, _) = data_provider.load_mnist(binary=True, standard_split=False)
(_,_), (_,_), (test_x,_) = data_tools.load_mnist('../../data')
(_,_), (_,_), (test_x_b,_) = data_tools.load_mnist_binary('../../data')
means = test_x.astype('float32')
means = numpy.clip(means,1e-10,(1-(1e-5)))
#means = numpy.random.uniform(size=(10000,784)).astype('float32') * 0 + 0.5
csl = CSL()
minibatches = test_x_b.reshape((1000,10,784)).astype('float32')
if 0:
# when means is a matrix of (N,D), representing only 1 chain
csl_fn = csl.get_CSL_fn_independent_Bernoulli_v2(means)
csl.compute_CSL_with_minibatches_one_chain(csl_fn, minibatches)
else:
# when means is a 3D tensor (N, K, D)
# When there are N chains, each chain having K samples of dimension D
chains = means.reshape(10,100,10,784)
csl_fn = csl.get_CSL_fn_independent_Bernoulli()
csl.compute_CSL_with_minibatches(csl_fn, minibatches, chains)
def visualize_mnist():
(train_X, train_Y), (valid_X, valid_Y), (test_X, test_Y) = data.load_mnist('../data')
design_matrix = train_X
images = design_matrix[0:2500, :]
channel_length = 28 * 28
to_visualize = images
image_data = tile_raster_images(to_visualize,
img_shape=[28,28],
tile_shape=[50,50],
tile_spacing=(2,2))
im_new = Image.fromarray(numpy.uint8(image_data))
im_new.save('samples_mnist.png')
os.system('eog samples_mnist.png')