search

hippke / wotan

Automagically remove trends from time-series data
MIT License
145 stars 21 forks source link

Add L1 (Koh, Kim, Boyd) #38

Open hippke opened 5 years ago

hippke commented 5 years ago

Recommended by: https://arxiv.org/pdf/1908.07151.pdf Example using cvxopt

n = y.size
e = np.ones((1, n))
D = scipy.sparse.spdiags(np.vstack((e, -2*e, e)), range(3), n-2, n)
x = cp.Variable(shape=n)
obj = cp.Minimize(0.5 * cp.sum_squares(y - x) + degree * cp.norm(D*x, 1))
prob = cp.Problem(obj).solve()