genrp 0.0.1.dev1

UNKNOWN

Python and C++ implementation of the generalized Rybicki Press algorithm for solving matrices of the form:

K_{ij} = sum_p a_p exp(-b_p |t_i - t_j|)

The method was developed by Sivaram Ambikasaran and you must cite his paper if you use this code in your work.

This interface allows complex bs and the parameters are specified as log-amplitudes, log-Q-factors, and frequencies. A frequency can be set to None if it is meant to be non-periodic.

A simple benchmark shows that this scales as O(N):

import time
import numpy as np
import matplotlib.pyplot as pl

from ess import GRPSolver

solver = GRPSolver(
    np.log([10.0, 5.0]),  # log-amplitudes
    np.log([0.1, 10.0]),  # log-Q-factors
    [None, 50.0],         # frequencies
)

N = 2**np.arange(5, 20)
times = np.empty((len(N), 3))

t = np.random.rand(np.max(N))
yerr = np.random.uniform(0.1, 0.2, len(t))
b = np.random.randn(len(t))

for i, n in enumerate(N):
    strt = time.time()
    solver.compute(t[:n], yerr[:n])
    times[i, 0] = time.time() - strt

    strt = time.time()
    solver.log_determinant
    times[i, 1] = time.time() - strt

    strt = time.time()
    solver.apply_inverse(b[:n])
    times[i, 2] = time.time() - strt