def mcfit(logpost,bounds,args=(),sampler=None,proposal=None,N=10000,burnin=1000,seed=None,outfile=None):
"""
Fit args using MCMC and return samples from the PDF.
"""
if sampler == None:
# sampler = MCSampler()
sampler = EnsembleSampler(100)
try:
bounds = np.array(bounds)
# if np.shape(bounds)[1] != 2:
# raise MCError("provide bounds on the parameter space")
if proposal == None:
proposal = (bounds[:,1]-bounds[:,0])/10.
return sampler.sample_pdf(logpost,bounds,proposal,N,burnin,args,outfile=outfile,seed=None)
except (MCError,ProposalErr): # as e:
print "MCMC exception raised with message: "+e.value
def mcfit(lnposteriorfn,p0,args=(),sampler=None,N=1000,burnin=0,outfile=None):
"""
Performs a (kitchen-sink) MCMC to sampler a provided posterior PDF
"""
#Required Arguments:
# `lnposteriorfn`
# a function that returns the value of posterior PDF
# at the position `p` when called
# `lnposteriorfn(p,args)` where `args` is the unpacked
# optional keyword argument of `mcfit`
# p0
#
#Optional Arguments:
# args=()
# sampler=None
# N=1000
# burnin=0
# outfile=None
#
#"""
p0 = np.array(p0)
if sampler == None:
sampler = EnsembleSampler(np.shape(p0)[0],np.shape(p0)[1],
lnposteriorfn,postargs=args,outfile=outfile)
pos = p0
state = None
if burnin > 0:
print "Running: first burn-in pass"
pos,prob,state = sampler.run_mcmc(pos,state,burnin/2)
pos,prob,state = sampler.clustering(pos,prob,state)
print "Running: second burn-in pass"
pos,prob,state = sampler.run_mcmc(pos,state,burnin/2)
pos,prob,state = sampler.clustering(pos,prob,state)
sampler.clear_chain()
print "Running: final Markov chain (%d links)"%(N)
sampler = EnsembleSampler(np.shape(p0)[0],np.shape(p0)[1],lnposteriorfn,
postargs=args,outfile=outfile)
sampler.run_mcmc(pos,state,N)
frac = np.mean(sampler.acceptance_fraction())
if frac < 0.1:
print "Warning: acceptance fraction < 10%"
else:
print "Acceptance fraction: %.3f"%frac
return sampler.chain,sampler.probability,frac
