def testNormal(plot = False, perc = 1e-3):
N = int(1e6)
# effective number of simulations:
iterations = int(1e2)
y = nrm.likelihood.simul(3,100)
eps = ABC.findEpsilon(y = y, iterations = N, prior = nrm.prior,
likelihood = nrm.likelihood,
dist = misc.euclidean_dist, perc = perc,
sum_statistics = misc.meanVar)
thetas_all = []
t0 = time.clock()
thetas, zs = ABC.sampler(y = y, iterations = iterations,
prior = nrm.prior,
likelihood = nrm.likelihood,
dist = misc.euclidean_dist,
tolerance = eps,
sum_statistics = misc.meanVar)
thetas_all.append(thetas)
t1 = time.clock()
if plot:
misc.plotDensity(thetas_all, meanPrior = 0, sigmaPrior = 100,
empMean = np.mean(y), sigma = 4, n = 100)
print(np.mean(thetas))
return(t1-t0)
def test_MA2_samplers(which = "basic", noisy = False,
generator = "MA(2)", naive = True,
exactBF = False):
# simulate observation
if generator == "MA(2)":
y = MA2.likelihood.simul(theta = [0.6,0.2], n = 100+2)
elif generator == "MA(1)":
y = MA1.likelihood.simul(theta = 0.6, n = 100+1)
else:
raise ValueError("Invalid name of the generator, "
"only MA(1) and MA(2) are allowed.")
# total number of simulations
N = int(1e6)
# expected proportion of accepted simulations
perc = 1e-3
# effective number of simulations:
iterations = int(N * perc)
if which == "MCMC":
eps = 60
thetas, zs = ABC.MCMCsampler(y = y, iterations = iterations,
prior = MA2.prior,
likelihood = MA2.likelihood,
dist = misc.euclidean_dist,
tolerance = eps,
sum_statistics = misc.autocov,
transKernel = MA2.transKernel)
misc.plot_thetas(thetas)
elif which == "basic":
eps = ABC.findEpsilon(y = y, iterations = N, prior = MA2.prior,
likelihood = MA2.likelihood,
dist = misc.euclidean_dist, perc = perc,
sum_statistics = misc.autocov)
thetas, zs = ABC.sampler(y = y, iterations = iterations,
prior = MA2.prior,
likelihood = MA2.likelihood,
dist = misc.euclidean_dist,
tolerance = eps,
sum_statistics = misc.autocov)
misc.plot_thetas(thetas)
elif which == "importance":
thetas, zs, weights = ABC.importanceSampler(y = y,
iterations = int(1e5),
prior = MA2.prior,
likelihood = MA2.likelihood,
proposal = MA2.proposal,
densKernel = MA2.densKernel,
bandwidth = 10,
sum_statistics = misc.autocov,
noisy = noisy)
out = np.sum([t * w for t,w in
zip(thetas, weights)],axis = 0)/np.sum(weights)
print("Evaluation of E[theta] using importance sampler gives: ", out)
elif which == "KernelMCMC":
thetas, zs = ABC.kernelMCMCSampler(y = y, iterations = int(1e4),
prior = MA2.prior,
likelihood = MA2.likelihood,
densKernel = MA2.densKernel,
bandwidth = 10,
transKernel = MA2.transKernel,
sum_statistics = misc.autocov,
dist = misc.euclidean_dist,
tolerance = 10, noisy = noisy)
misc.plot_thetas(thetas)
elif which == "modelChoiceSampler":
eps = ABCv.findEpsilon(y = y, iterations = N, prior = MA2.prior,
likelihood = MA2.likelihood,
dist = misc.euclidean_dist_multi,
perc = np.array([0.01,0.001, 0.0001, 0.00001]),
sum_statistics = misc.autocov)
if naive:
evs = []
for e in eps:
thetas, zs, ms = ABC.modelChoiceSampler(y = y, iterations = iterations,
paramPriors = [MA1.prior, MA2.prior],
likelihoods = [MA1.likelihood, MA2.likelihood],
modelPrior = misc.twoModelsPrior,
dist = misc.euclidean_dist,
tolerance = 7,
sum_statistics = misc.autocov)
ev0 = len([m for m in ms if m == 0])/len(ms)
evs.append(ev0)
else:
evs = []
for e in eps:
ms = ABCv.modelChoiceSampler(y = y, iterations = N,
paramPriors = [MA1.prior, MA2.prior],
likelihoods = [MA1.likelihood, MA2.likelihood],
modelPrior = misc.twoModelsPrior,
dist = misc.euclidean_dist_multi,
tolerance = e,
sum_statistics = misc.autocov)
ev0 = ms[0]/(ms[0]+ms[1])
evs.append(ev0)
misc.plotMAModSel(evs)
if exactBF:
print("Finding actual Bayes Factor...\n")
ev0exact = bf.MALogBayesFactor(y, resolution=(100,200))
for e in evs:
misc.printModelChoice("MA(1)", "MA(2)", e, ev0exact)
else:
raise ValueError("Invalid name of tested model.")
