def test_entropyWish(self): ''' Verify that (wishart) entropy is same for this object and Wishart object ''' Hself = self.distr.entropyWish() wishDistr = WishartDistr(v=self.distr.dF, invW=self.distr.invW) Hwish = wishDistr.get_entropy() assert np.allclose( Hself, Hwish)
def test_entropyWish(self):
''' Verify that (wishart) entropy is same for this object and Wishart object
'''
Hself = self.distr.entropyWish()
wishDistr = WishartDistr(v=self.distr.dF, invW=self.distr.invW)
Hwish = wishDistr.get_entropy()
assert np.allclose(Hself, Hwish)
def test_post_update_soVB(self, rho=0.375):
distrA = copy.deepcopy(self.distr)
distrB = WishartDistr(distrA.v, invW=np.eye(distrA.D))
self.distr.post_update_soVB(rho, distrB)
assert self.distr.v == rho * distrA.v + (1 - rho) * distrB.v
assert np.allclose(self.distr.invW,
rho * distrA.invW + (1 - rho) * distrB.invW)
class TestWishart(object):
def setUp(self):
self.v = 4
self.invW = np.eye(2)
self.distr = WishartDistr(v=self.v, invW=self.invW)
def test_dimension(self):
assert self.distr.D == self.invW.shape[0]
def test_cholinvW(self):
cholinvW = self.distr.cholinvW()
assert np.allclose(np.dot(cholinvW, cholinvW.T), self.distr.invW)
def test_expected_covariance_matrix(self):
CovMat = self.distr.ECovMat()
MyCovMat = self.invW / (self.v - self.distr.D - 1)
print MyCovMat, CovMat
assert np.allclose(MyCovMat, CovMat)
def test_post_update_soVB(self, rho=0.375):
distrA = copy.deepcopy(self.distr)
distrB = WishartDistr(distrA.v, invW=np.eye(distrA.D) )
self.distr.post_update_soVB(rho, distrB)
assert self.distr.v == rho*distrA.v + (1-rho)*distrB.v
assert np.allclose(self.distr.invW, rho*distrA.invW + (1-rho)*distrB.invW)
def test_entropy_posterior_gets_smaller(self, N=1):
PRNG = np.random.RandomState(seed=8675309)
for trial in range(3):
X = PRNG.randn(N, self.distr.D)
xxT = np.dot(X.T, X)
SS = SuffStatBag(K=1, D=self.distr.D)
SS.setField('N', [N], dims='K')
SS.setField('xxT', [xxT], dims=('K','D','D'))
postD = self.distr.get_post_distr(SS, 0)
assert postD.D == self.distr.D
Hpost = postD.get_entropy()
Hprior = self.distr.get_entropy()
print 'Prior %.3g, Post %.3g' % (Hprior, Hpost)
print self.distr.invW
print postD.invW
assert Hpost < Hprior
def CreateWithPrior(cls, inferType, priorArgDict, Data):
D = Data.dim
if inferType == 'EM':
obsPrior = None
return cls(inferType,
D,
obsPrior,
min_covar=priorArgDict['min_covar'])
else:
obsPrior = WishartDistr.CreateAsPrior(priorArgDict, Data)
return cls(inferType, D, obsPrior)
def CreateWithAllComps(cls, oDict, obsPrior, compDictList):
if 'min_covar' in oDict:
mc = oDict['min_covar']
self = cls(oDict['inferType'], obsPrior=obsPrior, min_covar=mc)
else:
self = cls(oDict['inferType'], obsPrior=obsPrior)
self.K = len(compDictList)
self.comp = [None for k in range(self.K)]
if self.inferType == 'EM':
for k in xrange(self.K):
self.comp[k] = ZMGaussDistr(**compDictList[k])
self.D = self.comp[k].D
elif self.inferType.count('VB') > 0:
for k in xrange(self.K):
self.comp[k] = WishartDistr(**compDictList[k])
self.D = self.comp[k].D
return self
class TestWishart(object):
def setUp(self):
self.v = 4
self.invW = np.eye(2)
self.distr = WishartDistr(v=self.v, invW=self.invW)
def test_dimension(self):
assert self.distr.D == self.invW.shape[0]
def test_cholinvW(self):
cholinvW = self.distr.cholinvW()
assert np.allclose(np.dot(cholinvW, cholinvW.T), self.distr.invW)
def test_expected_covariance_matrix(self):
CovMat = self.distr.ECovMat()
MyCovMat = self.invW / (self.v - self.distr.D - 1)
print MyCovMat, CovMat
assert np.allclose(MyCovMat, CovMat)
def test_post_update_soVB(self, rho=0.375):
distrA = copy.deepcopy(self.distr)
distrB = WishartDistr(distrA.v, invW=np.eye(distrA.D))
self.distr.post_update_soVB(rho, distrB)
assert self.distr.v == rho * distrA.v + (1 - rho) * distrB.v
assert np.allclose(self.distr.invW,
rho * distrA.invW + (1 - rho) * distrB.invW)
def test_entropy_posterior_gets_smaller(self, N=1):
PRNG = np.random.RandomState(seed=8675309)
for trial in range(3):
X = PRNG.randn(N, self.distr.D)
xxT = np.dot(X.T, X)
SS = SuffStatBag(K=1, D=self.distr.D)
SS.setField('N', [N], dims='K')
SS.setField('xxT', [xxT], dims=('K', 'D', 'D'))
postD = self.distr.get_post_distr(SS, 0)
assert postD.D == self.distr.D
Hpost = postD.get_entropy()
Hprior = self.distr.get_entropy()
print 'Prior %.3g, Post %.3g' % (Hprior, Hpost)
print self.distr.invW
print postD.invW
assert Hpost < Hprior
def setUp(self): self.v = 4 self.invW = np.eye(2) self.distr = WishartDistr(v=self.v, invW=self.invW)
def setUp(self):
self.v = 4
self.invW = np.eye(2)
self.distr = WishartDistr(v=self.v, invW=self.invW)
