use_gpu = int(options.gpu)
verbosity = int(options.verbose)
N = int(1e5) # n data points per component
K = 2 # ndim
ncomps = 3 # n mixture components
npr.seed(datetime.now().microsecond)
true_labels, data = generate_data(n=N, k=K, ncomps=ncomps)
data = data - data.mean(0)
data = data / data.std(0)
#import pdb
#pdb.set_trace()
print "use_gpu=" + str(use_gpu)
mcmc = DPNormalMixture(data,
ncomp=3,
gpu=use_gpu,
verbose=verbosity,
parallel=options.parallel) #, mu0=mu0)
mcmc.sample(200, nburn=0)
print mcmc.mu[-1]
bem = BEM_DPNormalMixture(mcmc, verbose=verbosity)
bem.optimize(maxiter=5)
print bem.mu
ident_mcmc = DPNormalMixture(bem, verbose=verbosity)
ident_mcmc.sample(100, nburn=0, ident=True)
print ident_mcmc.weights[-1]
print ident_mcmc.mu[-1]
def _fit(self, fcmdata, verbose=False):
"""
fit the mixture model to the data
use get_results() to get the fitted model
"""
pnts = fcmdata.view().copy()
self.m = pnts.mean(0)
self.s = pnts.std(0)
self.data = (pnts - self.m) / self.s
if len(self.data.shape) == 1:
self.data = self.data.reshape((self.data.shape[0], 1))
if len(self.data.shape) != 2:
raise ValueError("pnts is the wrong shape")
self.n, self.d = self.data.shape
if self._ref is not None:
self.ident = True
self._load_ref_at_fit(pnts)
if self.prior_mu is not None:
self._load_mu_at_fit()
if self.prior_sigma is not None:
self._load_sigma_at_fit()
if self.seed is not None:
seed(self.seed)
else:
from datetime import datetime
seed(datetime.now().microsecond)
#TODO move hyperparameter settings here
if self.type.lower() == 'bem':
self.cdp = BEM_DPNormalMixture(self.data,
ncomp=self.nclusts,
gamma0=self.gamma0,
m0=self.m0,
nu0=self.nu0,
Phi0=self.Phi0,
e0=self.e0,
f0=self.f0,
mu0=self._prior_mu,
Sigma0=self._prior_sigma,
weights0=self._prior_pi,
alpha0=self.alpha0,
gpu=self.device,
parallel=self.parallel,
verbose=verbose)
self.cdp.optimize(self.niter)
else:
self.cdp = DPNormalMixture(self.data,
ncomp=self.nclusts,
gamma0=self.gamma0,
m0=self.m0,
nu0=self.nu0,
Phi0=self.Phi0,
e0=self.e0,
f0=self.f0,
mu0=self._prior_mu,
Sigma0=self._prior_sigma,
weights0=self._prior_pi,
alpha0=self.alpha0,
gpu=self.device,
parallel=self.parallel,
verbose=verbose)
self.cdp.sample(niter=self.niter,
nburn=self.burnin,
thin=1,
ident=self.ident)
if self.last is None:
self.last = self.niter
self._run = True #we've fit the mixture model
return self.get_results()
