def check(train_images):
global patch_dim, image_dim
WB, zca_white, avg = lt.load_features('../LinearDecoder/theta.txt')
w = WB[0][0]
b = WB[0][1]
print 'zca_white:', zca_white.shape
print 'w:', w.shape
print 'b:', b.shape
print 'avg:', avg.shape
conv_images = train_images[:, :, :, :8]
convolved_features = convolve(patch_dim, 400, conv_images, w, b, zca_white, avg)
for _ in range(1000):
feature_num = random.randint(0, 399)
image_num = random.randint(0, 7)
image_row = random.randint(0, image_dim - patch_dim)
image_col = random.randint(0, image_dim - patch_dim)
patch = conv_images[image_row:image_row + patch_dim, image_col:image_col + patch_dim, :, image_num]
patch = np.concatenate((patch[:, :, 0].flatten(), patch[:, :, 1].flatten(), patch[:, :, 2].flatten()))
patch = patch.reshape((patch.size, 1))
patch = patch - avg
patch = zca_white.dot(patch)
print patch.shape
features = w.dot(patch) + b.reshape((b.shape[0], 1))
print features.shape
features = sigmoid(features)
print features.shape
print '*******************'
print features.shape
print convolved_features.shape
print feature_num, image_num, image_row, image_col
if np.abs(features[feature_num, 0] - convolved_features[feature_num, image_num, image_row, image_col]) > 1e-9:
print('Convolved feature does not match activation from autoencoder\n')
print('Feature Number : %d\n', feature_num)
print('Image Number : %d\n', image_num)
print('Image Row : %d\n', image_row)
print('Image Column : %d\n', image_col)
print('Convolved feature : %0.5f\n', convolved_features[feature_num, image_num, image_row, image_col])
print('Sparse AE feature : %0.5f\n', features[feature_num, 0])
print('Convolved feature does not match activation from autoencoder')
print 'Congratulations! Your convolution code passed the test.'
test_matrix = np.arange(64).reshape(8, 8)
expected_matrix = np.array([[np.mean(test_matrix[0:4, 0:4]), np.mean(test_matrix[0:4, 4:8])],
[np.mean(test_matrix[4:8, 0:4]), np.mean(test_matrix[4:8, 4:8])]])
test_matrix = np.reshape(test_matrix, (1, 1, 8, 8))
pooled_features = mean_pool(4, test_matrix)
if not (pooled_features == expected_matrix).all():
print "Pooling incorrect"
print "Expected matrix"
print expected_matrix
print "Got"
print pooled_features
print 'Congratulations! Your pooling code passed the test.'
def cnn(images, num_images, w, b, zca_white, avg):
global patch_dim, pool_dim
num_features = w.shape[0]
image_dim = images.shape[0]
step_size = 10
steps = int(math.ceil(num_features/step_size))
pool_size = math.floor((image_dim-patch_dim+1)/pool_dim)
pooled_features = np.zeros((num_features, num_images, pool_size, pool_size))
for i in range(steps):
feature_start = i*step_size;
feature_end = (i+1)*step_size;
wt = w[feature_start:feature_end, :]
bt = b[feature_start:feature_end]
convolved_features_this = convolve(patch_dim, step_size, images, wt, bt, zca_white, avg)
pooled_features_this = mean_pool(pool_dim, convolved_features_this)
pooled_features[feature_start:feature_end, :, :, :] = pooled_features_this
return pooled_features