def test_local_update(N, beta, family):
M = beta.shape[0]
X = np.random.random((N, M))
y = np.random.random(N) > 0.4
u = np.zeros(M)
z = np.random.random(M)
rho = 1e7
def create_local_gradient(func):
def wrapped(beta, X, y, z, u, rho):
return func(beta, X, y) + rho * (beta - z + u)
return wrapped
def create_local_f(func):
def wrapped(beta, X, y, z, u, rho):
return func(beta, X,
y) + (rho / 2) * np.dot(beta - z + u, beta - z + u)
return wrapped
f = create_local_f(family.pointwise_loss)
fprime = create_local_gradient(family.pointwise_gradient)
result = local_update(X, y, beta, z, u, rho, f=f, fprime=fprime)
assert np.allclose(result, z, atol=2e-3)
def test_local_update(N, beta):
M = beta.shape[0]
X = np.random.random((N, M))
y = np.random.random(N) > 0.4
u = np.zeros(M)
z = np.random.random(M)
rho = 1e6
result = local_update(X, y, beta, z, u, rho)
assert np.allclose(result, z, atol=2e-3)