test_robustpca.py

from robustpca import *
from numpy.random import randn
from numpy.random import shuffle
import numpy as np
from numpy.testing import assert_almost_equal

# initalize a basis of rank 10
Basis = randn(5, 100)
W = randn(300, 5)
TrueMat = np.dot(W, Basis)
# initalize a sparse matrix
sparse_ratio = 0.05
sparse_mag = 10
E = randn(TrueMat.size) * sparse_mag
idx = np.arange(E.size)
shuffle(idx)
E[idx[int(sparse_ratio * E.size):]] = 0
E = E.reshape(TrueMat.shape)
# calculate the observation
Observed = TrueMat + E  # 1000 x 100


def test_singular_value_thresholding():
    return
    A_, E_ = singular_value_thresholding(
        Observed.T, lmbda=0.1, tau=1e3, maxiter=100, verbose=0)
    assert (
        np.sqrt(((A_ - TrueMat.T)**2).sum()) / np.sqrt((A_**2).sum())) < .01
    assert (np.abs(E_ - E.T).sum() / np.abs(E).sum()) < .01


def test_augmented_largrange_multiplier():
    return
    A_, E_ = augmented_largrange_multiplier(
        Observed.T, lmbda=.1, inexact=True, verbose=0)
    assert_almost_equal(A_, TrueMat.T, decimal=2)
    assert_almost_equal(E_, E.T, decimal=2)
    A_, E_ = augmented_largrange_multiplier(
        Observed.T, lmbda=.1, inexact=False, verbose=0)
    assert_almost_equal(A_, TrueMat.T, decimal=2)
    assert_almost_equal(E_, E.T, decimal=2)


def test_accelerate_proximal_gradient():
    return
    A_, E_ = accelerate_proximal_gradient(Observed.T, lmbda=.1, verbose=0)
    assert_almost_equal(A_, TrueMat.T, decimal=2)
    assert_almost_equal(E_, E.T, decimal=2)


def test_alternating_direction_method_of_multipliers():
    A_, E_ = alternating_direction_method_of_multipliers(
        Observed.T, lmbda=.1, verbose=0)
    assert_almost_equal(A_, TrueMat.T, decimal=2)
    assert_almost_equal(E_, E.T, decimal=2)