def updateTau(self): #perform max_iter Gibbs updates
iter = 0
self.lp = self.logPosterior(self.gamma_store[0,:],self.tau,self.eta_store[0,:])
self.lp_star = self.lp
self.tau_star = np.copy(self.tau)
self.tau_store[iter,]=np.copy(self.tau)
while (iter < self.max_iter):
nchange = sampletau.sample_tau(self.tau, self.gamma_store[iter,:], self.eta_store[iter,:], self.variants)
#nchange = self.sampleTau(self.gamma_star,self.eta_star)
self.ll = self.logLikelihood(self.gamma_store[iter,:],self.tau,self.eta_store[iter,:])
self.lp = self.logPosterior(self.gamma_store[iter,:],self.tau,self.eta_store[iter,:])
if (self.lp > self.lp_star):
self.tau_star = np.copy(self.tau)
self.lp_star = self.lp
self.tau_store[iter,]=np.copy(self.tau)
self.ll_store[iter]=self.ll
if (iter % 10 == 0):
logging.info('Gibbs Iter %d, no. changed = %d, nll = %f'%(iter,nchange,self.lp))
sys.stdout.flush()
iter = iter + 1
self.updateTauIndices()
def update(self): #perform max_iter Gibbs updates
iter = 0
self.ll = self.logLikelihood(self.gamma,self.tau,self.eta)
self.lp = self.logPosterior(self.gamma,self.tau,self.eta)
self.storeStarState(iter)
while (iter < self.max_iter):
self.sampleMu(self.tau,self.gamma,self.eta)
self.sampleGamma()
#nchange = self.sampleTau()
nchange = sampletau.sample_tau(self.tau, self.gamma, self.eta, self.variants)
self.sampleEta()
self.ll = self.logLikelihood(self.gamma,self.tau,self.eta)
self.lp = self.logPosterior(self.gamma,self.tau,self.eta)
self.ll_store[iter] = self.ll
if(self.lp > self.lp_star):
self.storeStarState(iter)
self.mu_store[iter,]=np.copy(self.mu)
self.tau_store[iter,]=np.copy(self.tau)
self.E_store[iter,]=np.copy(self.E)
self.eta_store[iter,] = np.copy(self.eta)
self.gamma_store[iter,] = np.copy(self.gamma)
if (iter % 10 == 0):
logging.info('Gibbs Iter %d, no. changed = %d, nlp = %f'%(iter,nchange,self.lp))
iter = iter + 1
self.updateTauIndices()
def updateTau(self): #perform max_iter Gibbs updates
iter = 0
self.lp = self.logPosterior(self.gamma_store[0,:],self.tau,self.eta_store[0,:])
self.lp_star = self.lp
self.tau_star = np.copy(self.tau)
self.tau_store[iter,]=np.copy(self.tau)
while (iter < self.max_iter):
nchange = sampletau.sample_tau(self.tau, self.gamma_store[iter,:], self.eta_store[iter,:], self.variants)
#nchange = self.sampleTau(self.gamma_star,self.eta_star)
self.ll = self.logLikelihood(self.gamma_store[iter,:],self.tau,self.eta_store[iter,:])
self.lp = self.logPosterior(self.gamma_store[iter,:],self.tau,self.eta_store[iter,:])
if (self.lp > self.lp_star):
self.tau_star = np.copy(self.tau)
self.lp_star = self.lp
self.tau_store[iter,]=np.copy(self.tau)
self.ll_store[iter]=self.ll
if (iter % 10 == 0):
logging.info('Gibbs Iter %d, no. changed = %d, nll = %f'%(iter,nchange,self.lp))
sys.stdout.flush()
iter = iter + 1
self.updateTauIndices()
def update(self): #perform max_iter Gibbs updates
iter = 0
self.ll = self.logLikelihood(self.gamma,self.tau,self.eta)
self.lp = self.logPosterior(self.gamma,self.tau,self.eta)
self.storeStarState(iter)
while (iter < self.max_iter):
self.sampleMu(self.tau,self.gamma,self.eta)
self.sampleGamma()
#nchange = self.sampleTau()
nchange = sampletau.sample_tau(self.tau, self.gamma, self.eta, self.variants)
self.sampleEta()
self.ll = self.logLikelihood(self.gamma,self.tau,self.eta)
self.lp = self.logPosterior(self.gamma,self.tau,self.eta)
self.ll_store[iter] = self.ll
if(self.lp > self.lp_star):
self.storeStarState(iter)
self.mu_store[iter,]=np.copy(self.mu)
self.tau_store[iter,]=np.copy(self.tau)
self.E_store[iter,]=np.copy(self.E)
self.eta_store[iter,] = np.copy(self.eta)
self.gamma_store[iter,] = np.copy(self.gamma)
if (iter % 10 == 0):
logging.info('Gibbs Iter %d, no. changed = %d, nlp = %f'%(iter,nchange,self.lp))
iter = iter + 1
self.updateTauIndices()
def sampleTauC(self, tau, variants, eta, gamma=None, epsilon=None):
if gamma is None:
gamma = self.gamma
if epsilon is None:
epsilon = self.epsilon
V = tau.shape[0]
#rework gamma matrix
gammaR = self.maskGamma(gamma, eta)
nchange = sampletau.sample_tau(tau, gammaR, epsilon, variants)
return nchange
def sampleTauC(self,tau,variants,eta,gamma=None,epsilon=None):
if gamma is None:
gamma = self.gamma
if epsilon is None:
epsilon = self.epsilon
V = tau.shape[0]
#rework gamma matrix
gammaR = self.maskGamma(gamma,eta)
nchange = sampletau.sample_tau(tau, gammaR, epsilon, variants)
return nchange
