if __name__ == "__main__":
numpy.random.seed(0)
train_images, T = load_mnist("training", 60000)
train_images -= train_images.mean(axis=0)
test_images, T2 = load_mnist("testing", 10000)
test_images -= train_images.mean(axis=0)
print "Dataset loaded"
train_filter = train_images[:10000]
train_classifier = train_images
label_classifier = T
n_filters = 196
estimator = SparseFiltering(n_filters=n_filters, maxfun=500,
verbose=True)
estimator.fit(train_filter)
X = estimator.predict(train_classifier)
X2 = estimator.predict(test_images)
X_mean = X.mean(axis=0)
X_std = X.std(axis=0) + 1e-8
X = scale_features(X, X_mean, X_std)
X2 = scale_features(X2, X_mean, X_std)
print "Transformed datasets"
test_classifier(X, label_classifier, X2, T2)
pylab.figure()
pylab.subplots_adjust(wspace=0.0, hspace=0.0)
for i in range(estimator.W_.shape[0]):
rows = int(numpy.sqrt(n_filters))
pstd = 3 * numpy.std(data)
data = numpy.fmax(numpy.fmin(data, pstd), -pstd) / pstd
data = (data + 1) * 0.4 + 0.1;
return data
images = normalize_data(images)
patch_width = 8
n_filters = 25
n_samples, n_rows, n_cols = images.shape
n_features = n_rows * n_cols
patches = [extract_patches_2d(images[i], (patch_width, patch_width),
max_patches=1000, random_state=i)
for i in range(n_samples)]
patches = numpy.array(patches).reshape(-1, patch_width * patch_width)
print("Dataset consists of %d samples" % n_samples)
estimator = SparseFiltering(n_filters=n_filters, maxfun=200, verbose=True)
estimator.fit(patches)
pylab.figure()
for i in range(estimator.W_.shape[0]):
rows = int(numpy.sqrt(n_filters))
cols = int(numpy.sqrt(n_filters))
pylab.subplot(rows, cols, i + 1)
pylab.imshow(estimator.W_[i].reshape(patch_width, patch_width),
cmap=pylab.cm.gray, interpolation="nearest")
pylab.xticks(())
pylab.yticks(())
pylab.show()