def main(testing=True):
images = mnist.load_images('../data/train-images-idx3-ubyte') # 784 x 60000
labels = mnist.load_labels('../data/train-labels-idx1-ubyte') # 60000 x 1
util.display_network(images[:,0:100]) # Show the first 100 images
visible_size = 28*28
hidden_size = 196
sparsity_param = 0.1
lamb = 3e-3
beta = 3
patches = images[:,0:10000]
theta = autoencoder.initialize_parameters(hidden_size, visible_size)
def sal(theta):
return autoencoder.sparse_autoencoder_loss(theta, visible_size, hidden_size, lamb,
sparsity_param, beta, patches)
x, f, d = scipy.optimize.fmin_l_bfgs_b(sal, theta, maxfun=400, iprint=1, m=20)
W1, W2, b1, b2 = autoencoder.unflatten(x, visible_size, hidden_size)
util.display_network(W1.T)
# - `hidden_size_l2` - layer 2 hidden size
# - `sparsity_param` - desired average activation of the hidden units
# (\\(\\rho\\) in the lecture notes)
# - `lamb` - weight decay parameter
# - `beta` - weight of sparsity penalty term
input_size = 28 * 28
num_classes = 10
hidden_size_l1 = 200
hidden_size_l2 = 200
sparsity_param = 0.1
lamb = 3e-3
beta = 3
# === Step 1: Load data from the MNIST database ===
train_data = mnist.load_images('../data/train-images-idx3-ubyte')
train_labels = mnist.load_labels('../data/train-labels-idx1-ubyte')
# For debugging purposes, reduce the size of the input data in order
# to speed up gradient checking. Here, we consider only the eight
# most-varying pixels of the images, and only the first 100 images.
if DEBUG:
input_size = 64
# only 100 datapoints
train_data = train_data[:, :100]
train_labels = train_labels[:100]
# only top input_size most-varying input elements (pixels)
indices = train_data.var(1).argsort()[-input_size:]
train_data = np.asfortranarray(train_data[indices, :])
# === Step 2: Train the first sparse autoencoder ===
# Here we define and initialise some constants which allow the code
# to be used more generally on any arbitrary input. We also
# initialise some parameters used for tuning the model.
input_size = 28 * 28
num_classes = 10
lamb = 1e-4
# === Step 1: Load data ===
#
# In this section, we load the input and output data. For softmax
# regression on MNIST pixels, the input data is the images, and the
# output data is the labels.
images = mnist.load_images('../data/train-images-idx3-ubyte')
labels = mnist.load_labels('../data/train-labels-idx1-ubyte')
input_data = images
# For debugging purposes, reduce the size of the input data in order
# to speed up gradient checking. Here, we consider only the eight
# most-varying pixels of the images, and only the first 100 images.
DEBUG = False
if DEBUG:
input_size = 8
# only 100 datapoints
input_data = input_data[:, :100]
labels = labels[:100]
# only top input_size most-varying input elements (pixels)
indices = input_data.var(1).argsort()[-input_size:]
input_data = np.asfortranarray(input_data[indices, :])
# - `hidden_size_l2` - layer 2 hidden size
# - `sparsity_param` - desired average activation of the hidden units
# (\\(\\rho\\) in the lecture notes)
# - `lamb` - weight decay parameter
# - `beta` - weight of sparsity penalty term
input_size = 28 * 28
num_classes = 10
hidden_size_l1 = 200
hidden_size_l2 = 200
sparsity_param = 0.1
lamb = 3e-3
beta = 3
# === Step 1: Load data from the MNIST database ===
train_data = mnist.load_images("../data/train-images-idx3-ubyte")
train_labels = mnist.load_labels("../data/train-labels-idx1-ubyte")
# For debugging purposes, reduce the size of the input data in order
# to speed up gradient checking. Here, we consider only the eight
# most-varying pixels of the images, and only the first 100 images.
if DEBUG:
input_size = 64
# only 100 datapoints
train_data = train_data[:, :100]
train_labels = train_labels[:100]
# only top input_size most-varying input elements (pixels)
indices = train_data.var(1).argsort()[-input_size:]
train_data = np.asfortranarray(train_data[indices, :])
# === Step 2: Train the first sparse autoencoder ===