def random_sample(self):
"""Return a [NOBJ, NDIM] numpy array sampling over NSAMPLE using inverse interim_prior
weights. Needed to compute a posterior object."""
ps = 1. / self.interim_prior
ps /= np.sum(ps, axis=1)[:, np.newaxis]
return np.array(
[self.data[i, pick_discrete(p)] for i, p in enumerate(ps)])
def draw_new_label(self, i):
# This is essentially Neal (2000) equation (3.6)
# Note that the p probabilities are unnormalized here, but pick_discrete will rescale them
# so that the total probability is 1.0. This normalization also captures the factors of
# b/(n-1+alpha) in Neal (2000).
# -1 is sentinel for "make a new cluster"
picked = pick_discrete(self.p[i] * np.append([1], self.nphi)) - 1
return picked
def draw_new_label(self, i):
# This is essentially Neal (2000) equation (3.6)
# Note that the p probabilities are unnormalized here, but pick_discrete will rescale them
# so that the total probability is 1.0. This normalization also captures the factors of
# b/(n-1+alpha) in Neal (2000).
# -1 is sentinel for "make a new cluster"
picked = pick_discrete(self.p[i]*np.append([1], self.nphi)) - 1
return picked
def update_latent_data(self):
# Update the latent "true" data in the case that the data is represented by a
# Pseudo-marginal samples or (TBD) means and Gaussian errors.
if isinstance(self._D, PseudoMarginalData):
for i, ph in enumerate(self.phi):
index = np.nonzero(self.label == i)[0]
data = self._D[index] # a PseudoMarginalData instance
# calculate weights for selecting a representative sample
ps = self.prior.like1(self.manip(data.data), ph) / data.interim_prior
ps /= np.sum(ps, axis=1)[:, np.newaxis] # think this line can go.
for j, p in enumerate(ps):
self.D[index[j]] = data.data[j, pick_discrete(p)]
# Need to update the r_i probabilities too since self.D changed.
# self.r_i = self.alpha * self.prior.pred(self.mD)
self.p[:, 0] = self.alpha * self.prior.pred(self.mD)
self.manip_needs_update = True
else:
pass # If data is already a numpy array, there's nothing to update.
def update_latent_data(self):
# Update the latent "true" data in the case that the data is represented by a
# Pseudo-marginal samples or (TBD) means and Gaussian errors.
if isinstance(self._D, PseudoMarginalData):
for i, ph in enumerate(self.phi):
index = np.nonzero(self.label == i)[0]
data = self._D[index] # a PseudoMarginalData instance
# calculate weights for selecting a representative sample
ps = self.prior.like1(self.manip(data.data),
ph) / data.interim_prior
ps /= np.sum(ps, axis=1)[:,
np.newaxis] # think this line can go.
for j, p in enumerate(ps):
self.D[index[j]] = data.data[j, pick_discrete(p)]
# Need to update the r_i probabilities too since self.D changed.
# self.r_i = self.alpha * self.prior.pred(self.mD)
self.p[:, 0] = self.alpha * self.prior.pred(self.mD)
self.manip_needs_update = True
else:
pass # If data is already a numpy array, there's nothing to update.
def random_sample(self):
"""Return a [NOBJ, NDIM] numpy array sampling over NSAMPLE using inverse interim_prior
weights. Needed to compute a posterior object."""
ps = 1./self.interim_prior
ps /= np.sum(ps, axis=1)[:, np.newaxis]
return np.array([self.data[i, pick_discrete(p)] for i, p in enumerate(ps)])
