def test_evidence(verbose=0,k=1):
"""
Compare the evidence estimated by Chib's method
with the variational evidence (free energy)
fixme : this one really takes time
"""
n=50
dim=2
x = nr.randn(n,dim)
x[:15] += 3
b = VBGMM(k,dim)
b.guess_priors(x)
b.initialize(x)
b.estimate(x)
vbe = b.evidence(x),
if verbose:
print 'vb: ',vbe,
niter = 1000
b = BGMM(k,dim)
b.guess_priors(x)
b.initialize(x)
b.sample(x,100)
w,cent,prec,pz = b.sample(x, niter=niter, mem=1)
bplugin = BGMM(k, dim, cent, prec, w)
bplugin.guess_priors(x)
bfchib = bplugin.bayes_factor(x, pz.astype(np.int), 1)
if verbose:
print ' chib:', bfchib
assert(bfchib>vbe)
def test_evidence(verbose=0, k=1):
# Compare the evidence estimated by Chib's method with the
# variational evidence (free energy) fixme : this one really takes
# time
n = 100
dim = 2
x = nr.randn(n, dim)
x[:30] += 3
show = 0
b = VBGMM(k, dim)
b.guess_priors(x)
b.initialize(x)
b.estimate(x)
vbe = (b.evidence(x),)
if verbose:
print "vb: ", vbe,
niter = 1000
b = BGMM(k, dim)
b.guess_priors(x)
b.initialize(x)
b.sample(x, 100)
w, cent, prec, pz = b.sample(x, niter=niter, mem=1)
bplugin = BGMM(k, dim, cent, prec, w)
bplugin.guess_priors(x)
bfchib = bplugin.Bfactor(x, pz.astype(np.int), 1)
if verbose:
print " chib:", bfchib
assert bfchib > vbe
def test_bgmm_gibbs(verbose=0):
# perform the estimation of a gmm using Gibbs sampling
n = 100
k = 2
dim = 2
niter = 1000
x = nr.randn(n, dim)
x[:30] += 2
b = BGMM(k, dim)
b.guess_priors(x)
b.initialize(x)
b.sample(x, 1)
w, cent, prec, pz = b.sample(x, niter, mem=1)
b.plugin(cent, prec, w)
z = pz[:, 0]
# fixme : find a less trivial test
assert z.max() + 1 == b.k
def test_gmm_bf(kmax=4, seed=1, verbose=1):
"""
perform a model selection procedure on a gmm
with Bayes factor estimations
Parameters
----------
kmax : range of values that are tested
seed=False: int, optionnal
If seed is not False, the random number generator is initialized
at a certain value
fixme : this one often fails. I don't really see why
"""
n=30
dim=2
if seed:
nr = np.random.RandomState([seed])
else:
import numpy.random as nr
x = nr.randn(n,dim)
niter = 1000
bbf = -np.infty
for k in range(1, kmax):
b = BGMM(k, dim)
b.guess_priors(x)
b.initialize(x)
b.sample(x, 100)
w, cent, prec, pz = b.sample(x, niter=niter, mem=1)
bplugin = BGMM(k, dim, cent, prec, w)
bplugin.guess_priors(x)
bfk = bplugin.bayes_factor(x, pz.astype(np.int))
if verbose:
print k, bfk
if bfk>bbf:
bestk = k
bbf = bfk
assert(bestk<3)
def test_gmm_bf(kmax=4, verbose=0):
# perform a model selection procedure on a gmm with Bayes factor
# estimations kmax : range of values that are tested
#
# fixme : this one often fails. I don't really see why
n = 30
dim = 2
x = nr.randn(n, dim)
# x[:30] += 2
niter = 3000
bbf = -np.infty
for k in range(1, kmax):
b = BGMM(k, dim)
b.guess_priors(x)
b.initialize(x)
b.sample(x, 100)
w, cent, prec, pz = b.sample(x, niter=niter, mem=1)
bplugin = BGMM(k, dim, cent, prec, w)
bplugin.guess_priors(x)
bfk = bplugin.Bfactor(x, pz.astype(np.int), 1)
if verbose:
print k, bfk
if bfk > bbf:
bestk = k
bbf = bfk
assert bestk < 3
