def test_imm_wnc2():
"""
Test the basic imm_wnc when null class is shrunk to 0
"""
n = 50
dim = 1
alpha = .5
g0 = 1.
x = np.random.rand(n, dim)
x[:.3*n] *= .2
x[:.1*n] *= .3
# instantiate
migmm = MixedIMM(alpha, dim)
migmm.set_priors(x)
migmm.set_constant_densities(null_dens=g0)
ncp = np.zeros(n)
# warming
migmm.sample(x, null_class_proba=ncp, niter=100, init=True)
g = np.linspace(0, 1, 101)
#sampling
like, pproba = migmm.sample(x, null_class_proba=ncp, niter=300)
assert like.min()>.1
assert like.max()<5.
assert (pproba==ncp).all()
def test_imm_wnc3():
"""
Test the basic imm_wnc when null class is of proba 1 (nothing is estimated)
"""
n = 50
dim = 1
alpha = .5
g0 = 1.
x = np.random.rand(n, dim)
x[:.3*n] *= .2
x[:.1*n] *= .3
# instantiate
migmm = MixedIMM(alpha, dim)
migmm.set_priors(x)
migmm.set_constant_densities(null_dens=g0)
ncp = np.ones(n)
# warming
migmm.sample(x, null_class_proba=ncp, niter=100, init=True)
g = np.linspace(0, 1, 101)
#sampling
like, pproba = migmm.sample(x, null_class_proba=ncp, niter=300)
assert (pproba==ncp).all()
def test_imm_wnc1():
"""
Test the basic imm_wnc, where the probaility under the null are random
"""
n = 50
dim = 1
alpha = .5
g0 = 1.
x = np.random.rand(n, dim)
x[:.3*n] *= .2
x[:.1*n] *= .3
# instantiate
migmm = MixedIMM(alpha, dim)
migmm.set_priors(x)
migmm.set_constant_densities(null_dens=g0)
ncp = np.random.rand(n)
# warming
migmm.sample(x, null_class_proba=ncp, niter=100, init=True)
g = np.reshape(np.linspace(0, 1, 101), (101, dim))
#sampling
like, pproba = migmm.sample(x, null_class_proba=ncp, niter=300,
sampling_points=g)
# the density should sum to 1
ds = 0.01*like.sum()
assert ds<1
assert ds>.8
assert np.sum(pproba>.5)>1
assert np.sum(pproba<.5)>1
def dpmm(gfc, alpha, g0, g1, dof, prior_precision, gf1, sub, burnin,
spatial_coords=None, nis=1000, co_clust=False, verbose=False):
"""
Apply the dpmm analysis to the data: python version
"""
from nipy.neurospin.clustering.imm import MixedIMM
dim = gfc.shape[1]
migmm = MixedIMM(alpha, dim)
migmm.set_priors(gfc)
migmm.set_constant_densities(null_dens=g0, prior_dens=g1)
migmm._prior_dof = dof
migmm._prior_scale = np.diag(prior_precision[0]/dof)
migmm._inv_prior_scale_ = [np.diag(dof*1./(prior_precision[0]))]
migmm.sample(gfc, null_class_proba=1-gf1, niter=burnin, init=False,
kfold=sub)
if verbose:
print 'number of components: ', migmm.k
#sampling
if co_clust:
like, pproba, co_clust = migmm.sample(
gfc, null_class_proba=1-gf1, niter=nis,
sampling_points=spatial_coords, kfold=sub, co_clustering=co_clust)
if verbose:
print 'number of components: ', migmm.k
return like, 1-pproba, co_clust
else:
like, pproba = migmm.sample(
gfc, null_class_proba=1-gf1, niter=nis,
sampling_points=spatial_coords, kfold=sub, co_clustering=co_clust)
if verbose:
print 'number of components: ', migmm.k
return like, 1-pproba
