def test_img1():
for n_features in [10, 20, 50]:
img = datasets.load_img1(n_features, 20)
assert img.shape[0] == n_features
assert img.shape[1] == 20
colors = ['#7fc97f', '#beaed4', '#fdc086']
from copt.total_variation import prox_tv2d, prox_tv1d_rows, prox_tv1d_cols
from copt import three_split, proximal_gradient
from copt.utils import Trace
from copt.datasets import load_img1
from scipy import misc
###############################################################
# Load an ground truth image and generate the dataset (A, b) as
#
# b = A ground_truth + noise ,
#
# where A is a random matrix. We will now load the ground truth image
img = load_img1()
n_rows, n_cols = img.shape
n_features = n_rows * n_cols
np.random.seed(0)
n_samples = n_features
# set L2 regularization (arbitrarily) to 1/n_samples
l2_reg = 1.0 / n_samples
A = np.random.uniform(-1, 1, size=(n_samples, n_features))
for i in range(A.shape[0]):
A[i] /= linalg.norm(A[i])
b = A.dot(img.ravel()) + 1.0 * np.random.randn(n_samples)
