def sample_model(self):
"""
Returns a list of model numbers, of length self.nbatch, obtained by sampling from a categorical distribution
with probabilities self.modelprior, and then perturbing with a uniform model perturbation kernel.
"""
models = [0] * self.nbatch
if self.nmodel > 1:
# Sample models from prior distribution
for i in range(self.nbatch):
models[i] = statistics.w_choice(self.margins_prev)
# perturb models
if len(self.dead_models) < self.nmodel - 1:
for i in range(self.nbatch):
u = rnd.uniform(low=0, high=1)
if u > self.modelKernel:
# sample randomly from other (non dead) models
not_available = set(self.dead_models[:])
not_available.add(models[i])
available_indexes = np.array(
list(set(range(self.nmodel)) - not_available))
rnd.shuffle(available_indexes)
perturbed_model = available_indexes[0]
models[i] = perturbed_model
return models[:]
def sampleTheModel(self):
ret = [0 for it in range(self.nbatch)]
if self.nmodel > 1:
for i in range(self.nbatch):
ret[i] = statistics.w_choice(range(self.nmodel),
self.margins_prev)
if (len(self.dead_models) < self.nmodel - 1):
#if(0):
# perturb models
for i in range(self.nbatch):
u = rnd.uniform(low=0, high=1)
if u > self.modelKernel:
# sample randomly from other (non dead) models
not_available = self.dead_models[:]
not_available.append(ret[i])
ss = set(not_available)
s = set(range(0, self.nmodel))
# print "perturbing model ", set( range(self.nmodel) ), ss, s-ss, list(s-ss), numpy.array( list(s-ss) )
ar = numpy.array(list(s - ss))
rnd.shuffle(ar)
perturbed_model = ar[0]
# print "perturbation ", ret[i], "->", perturbed_model
ret[i] = perturbed_model
return ret[:]
def sampleTheModel(self):
ret = [0 for it in range(self.nbatch)]
if self.nmodel > 1:
for i in range(self.nbatch):
ret[i] = statistics.w_choice( range(self.nmodel), self.margins_prev )
if( len(self.dead_models) < self.nmodel-1 ):
#if(0):
# perturb models
for i in range(self.nbatch):
u = rnd.uniform(low=0, high=1)
if u > self.modelKernel:
# sample randomly from other (non dead) models
not_available = self.dead_models[:]
not_available.append(ret[i])
ss = set( not_available )
s = set( range(0,self.nmodel) )
# print "perturbing model ", set( range(self.nmodel) ), ss, s-ss, list(s-ss), numpy.array( list(s-ss) )
ar = numpy.array( list(s-ss) )
rnd.shuffle( ar )
perturbed_model = ar[0]
# print "perturbation ", ret[i], "->", perturbed_model
ret[i] = perturbed_model
return ret[:]
def sampleTheModelFromPrior(self):
ret = [0 for it in range(self.nbatch)]
if self.nmodel > 1:
for i in range(self.nbatch):
ret[i] = statistics.w_choice( range(self.nmodel), self.modelprior )
return ret[:]
def sampleTheModelFromPrior(self):
ret = [0 for it in range(self.nbatch)]
if self.nmodel > 1:
for i in range(self.nbatch):
ret[i] = statistics.w_choice( range(self.nmodel), self.modelprior )
return ret[:]
def sample_model_from_prior(self):
"""
Returns a list of model numbers, of length self.nbatch, drawn from a categorical distribution with probabilities
self.modelprior
"""
models = [0] * self.nbatch
if self.nmodel > 1:
for i in range(self.nbatch):
models[i] = statistics.w_choice(self.modelprior)
return models