matthieubulte

using LinearAlgebra
using BenchmarkTools
using SparseArrays

# To exploit the sparsity of the matrix as much as possible, pre-compile the multiplication operation
# to skip the matrix entries look-up. Since the matrix is sparse this might generate only a few operations
# which are all inlined.
# Note that this only works with small matrices since otherwise the size of the function will just be to
# large.
# Note that this could also be avoided, but this is another topic.

matthieubulte

matthieubulte

matthieubulte

matthieubulte

matthieubulte

matthieubulte

matthieubulte

import multiprocessing as mp
import numpy as np

def multiprocessing(f, data, nproc):
    chunks = np.array_split(data, nproc)
    with mp.Pool(nproc) as ex:
        res = ex.map(f, chunks)
    return np.concatenate(list(res))

def dosomething(xs):

matthieubulte

matthieubulte

def smc(M, steps, prior, potential, proposal_kernel, correction_steps=1, ess_ratio=2):
    """
    smc(M, steps, prior, potential, proposal_kernel, correction_steps=1, ess_ratio=2)
    
    Creates a particle approximation of a probability distribution using the Sequential
    Monte Carlo (SMC) algorithm with Markov Chain Monte Carlo (MCMC) correction steps using
    the Metropolis-Hastings algorithm.
    
    Parameters