def joint_density(df_l, verbose=False):
df = df_l
val = -sum((wishart.logpdf(x, df=df, scale=mean_wishart / df)
for x in sample_of_matrices))
if verbose:
print df, ':', val
return val
def log_likelihood(self, params_opt, node_name):
params_node = self.get_node_params(params_opt, node_name)
m_sigma = self.calculate_sigma(params_node)
if node_name in self.m_cov.keys():
m_cov = self.m_cov[node_name]
else:
m_cov = self.get_matrices(params_node, 'Cov')
df = sum([p.shape[0] for _, p in self.m_profiles.items()])
w = wishart.logpdf(m_cov, df=df, scale=m_sigma / df)
return w
def log_likelihood(self, params):
""" Likelihood of data """
# m_sigma = self.calculate_sigma(params)
# w = 0
# for p in self.m_profiles:
# w += multivariate_normal.logpdf(p, np.zeros(p.shape), m_sigma)
m_sigma = self.calculate_sigma(params)
df = self.m_profiles.shape[0]
w = wishart.logpdf(self.m_cov, df=df, scale=m_sigma / df)
return w
def test_log_pdf(self, dtype_dof, dtype, degrees_of_freedom, random_state,
scale_is_samples, rv_is_samples, num_data_points, num_samples, broadcast):
# Create positive semi-definite matrices
rv = make_spd_matrices_4d(num_samples, num_data_points, degrees_of_freedom, random_state=random_state)
if broadcast:
scale = make_spd_matrix(n_dim=degrees_of_freedom, random_state=random_state)
else:
scale = make_spd_matrices_4d(num_samples, num_data_points, degrees_of_freedom, random_state=random_state)
degrees_of_freedom_mx = mx.nd.array([degrees_of_freedom], dtype=dtype_dof)
degrees_of_freedom = degrees_of_freedom_mx.asnumpy()[0] # ensures the correct dtype
scale_mx = mx.nd.array(scale, dtype=dtype)
if not scale_is_samples:
scale_mx = add_sample_dimension(mx.nd, scale_mx)
scale = scale_mx.asnumpy()
rv_mx = mx.nd.array(rv, dtype=dtype)
if not rv_is_samples:
rv_mx = add_sample_dimension(mx.nd, rv_mx)
rv = rv_mx.asnumpy()
is_samples_any = scale_is_samples or rv_is_samples
if broadcast:
scale_np = np.broadcast_to(scale, rv.shape)
else:
n_dim = 1 + len(rv.shape) if is_samples_any and not rv_is_samples else len(rv.shape)
scale_np = numpy_array_reshape(scale, is_samples_any, n_dim)
rv_np = numpy_array_reshape(rv, is_samples_any, degrees_of_freedom)
r = []
for s in range(num_samples):
a = []
for i in range(num_data_points):
a.append(wishart.logpdf(rv_np[s][i], df=degrees_of_freedom, scale=scale_np[s][i]))
r.append(a)
log_pdf_np = np.array(r)
var = Wishart.define_variable(shape=rv.shape[1:], dtype=dtype, rand_gen=None).factor
variables = {var.degrees_of_freedom.uuid: degrees_of_freedom_mx, var.scale.uuid: scale_mx,
var.random_variable.uuid: rv_mx}
log_pdf_rt = var.log_pdf(F=mx.nd, variables=variables)
assert np.issubdtype(log_pdf_rt.dtype, dtype)
assert is_sampled_array(mx.nd, log_pdf_rt) == is_samples_any
if is_samples_any:
assert get_num_samples(mx.nd, log_pdf_rt) == num_samples, (get_num_samples(mx.nd, log_pdf_rt), num_samples)
assert np.allclose(log_pdf_np, log_pdf_rt.asnumpy())
def test_CovUnconstrainedCholeskyWishartReg(seeded_rng):
cov = CovUnconstrainedCholeskyWishartReg(size=m)
L = cov.L.numpy()
cov_np = L @ L.T
logdet_np, sinv_np, sinvx_np = logdet_sinv_np(X, cov_np)
assert_allclose(logdet_np, cov.logdet, rtol=rtol)
assert_allclose(sinv_np, cov.solve(eye), rtol=rtol)
assert_allclose(sinvx_np, cov.solve(X_tf), rtol=rtol)
# now compute the regularizer
reg = wishart.logpdf(cov_np, df=m + 2, scale=1e10 * np.eye(m))
assert_allclose(reg, cov.logp, rtol=rtol)
def loglike_GW(obs1, obs2, scale, df, mean, scaler):
like_log = 0
# Wishart loglike
like_log += wishart.logpdf(obs2, df, scale)
# MVN loglike
# scaler the precision first
precision = []
for i in range(n_factor):
precision.append([])
for j in range(n_factor):
temp = obs2[i][j] * scaler
precision[i].append(temp)
like_log += loglike_MVN(obs1, mean, precision)
return like_log
def loglike_GW(obs1, obs2, scale, df, mean, scaler):
like_log = 0
# Wishart loglike
like_log += wishart.logpdf(obs2, df, scale)
# MVN loglike
# scaler the precision first
precision = []
for i in range(n_factor):
precision.append([])
for j in range(n_factor):
temp = obs2[i][j] * scaler
precision[i].append(temp)
like_log += loglike_MVN(obs1, mean, precision)
return like_log
def loglike_GW(obs1, obs2, scale, df, mean, scaler): like_log = 0 # Wishart loglike like_log += wishart.logpdf(obs2, df, scale) # MVN loglike # scaler the precision first precision = [] for i in range(n_factor): precision.append([]) for j in range(n_factor): temp = obs2[i][j] * scaler precision[i].append(temp) like_log += loglike_MVN(obs1, mean, precision) ## NOTE: ''' cov = inv(precision) like_log += multivariate_normal.logpdf(obs1, mean, cov) ''' return like_log
def loglike_GW(obs1, obs2, scale, df, mean, scaler):
like_log = 0
# Wishart loglike
like_log += wishart.logpdf(obs2, df, scale)
# MVN loglike
# scaler the precision first
precision = []
for i in range(n_factor):
precision.append([])
for j in range(n_factor):
temp = obs2[i][j] * scaler
precision[i].append(temp)
like_log += loglike_MVN(obs1, mean, precision)
## NOTE:
'''
cov = inv(precision)
like_log += multivariate_normal.logpdf(obs1, mean, cov)
'''
return like_log
def getLogPosteriorProbability(self, latent_z, n_cluster_samples, params,
normal_insts):
n_sample = len(latent_z)
n_i = [None] * len(n_cluster_samples)
for idx, c in n_cluster_samples.items():
n_i[idx] = c
n_i = np.array(n_i)
tmp = np.log(self.alpha) * self.n_cluster + gammaln(n_i).sum()
p_s = tmp - logRisingFact(self.alpha, n_sample)
posterior = p_s
for k in range(self.n_cluster):
cov = np.linalg.inv(self.beta * params[k][1])
p_mean = multivariate_normal(self.u0, cov).logpdf(params[k][0])
p_covar = wishart.logpdf(params[k][1], self.new, self.covar)
posterior += p_mean + p_covar
sample_k = self.getSampleK(k, latent_z)
posterior += normal_insts[k].logpdf(sample_k).sum()
return posterior
def adjust_treemix_df(wishart_df, starting_tree):
cov = make_covariance(starting_tree)
lmax = wishart.logpdf(wishart, scale=wishart / wishart_df, df=wishart_df)
def logpdf(x, v, n):
correction = (n * v.shape[0] / 2.)*np.log(2) + \
multigammaln(n/2, v.shape[0])
return [np.sum(wishart.logpdf(x, n, v)) + correction]
def propose(self):
vx = wishart.rvs(self.df, self.cx)
vy = wishart.rvs(self.df, self.cy)
fb = wishart.logpdf(vx, self.df, self.cx) + wishart.logpdf(vy, self.df, self.cy) - \
( wishart.logpdf(self.cx, self.df, vx) + wishart.logpdf(self.cy, self.df, vy) )
return AlignedNormal2D(cx=vx, cy=vy), fb
def compute_prior(self):
return wishart.logpdf(self.c, self.df, numpy.eye(self.df))
def logwishart(covariance,mean,df):
if isinstance(covariance, int) or isinstance(covariance, float):
return chi2.logpdf(covariance, scale=mean/df, df=df)
return scip_wishart.logpdf(covariance, scale=mean/df, df=df)
def wishart_distance(a,b):
try:
ans=(wishart.logpdf(a, df=df, scale=b/df)+wishart.logpdf(b, df=df, scale=a/df))/2
except np.linalg.LinAlgError as e:
ans=np.NaN
return ans
def get_average_deviation(scale, df, reps=1000):
sims=[wishart.rvs(scale=scale/df, df=df) for _ in xrange(reps)]
wishs=[wishart.logpdf(cov, df=df, scale=scale/df) for cov in sims]
dists=[dist(cov, scale) for cov in sims]
return (min(wishs), max(wishs)),(min(dists),max(dists))
'''
# calculate likelihood manually, following https://en.wikipedia.org/wiki/Wishart_distribution
# and https://math.stackexchange.com/questions/2803164/degrees-of-freedom-in-a-wishart-distribution
# and /Users/momchil/anaconda3/lib/python3.7/site-packages/semopy/stats.py: obj_mlw()
# and /Users/momchil/anaconda3/lib/python3.7/site-packages/semopy/stats.py: calc_bic()
# and /Users/momchil/anaconda3/lib/python3.7/site-packages/semopy/stats.py: calc_likelihood()
model.update_matrices(model.param_vals)
S = model.mx_cov # empirical covariance matrix
sigma, _ = model.calc_sigma() # model covariance matrix
p = sigma.shape[0] # # of variables
n = model.mx_data.shape[
0] # dof = # of data points (b/c Wishart is generalization of Chi2 to MVN => sum of n squared i.i.d. MVNs)
k = len(model.param_vals) # # of free params
loglik = wishart.logpdf(S, df=n, scale=sigma)
bic = k * np.log(n) - 2 * loglik
lme = -0.5 * bic
logliks[i, j] = loglik
ks[i, j] = k
ns[i, j] = n
bics[i, j] = bic
lmes[i, j] = lme
print('subj=', i, ' mod=', j, filepath, ' k=', k, ' n=', n,
' loglik=', logliks[i, j])
print(model.mx_cov.shape)
print(lmes.shape)
def propose(self):
val = wishart.rvs(self.df, self.c)
fb = wishart.logpdf(val, self.df, self.c) - wishart.logpdf(self.c, self.df, val)
return FreeNormal2D(c=val), fb
