Created on 01.02.2010
@author: floha
'''
import numpy as np
from scipy import stats
from matplotlib import pyplot
from fit import sampler, mcmc
print "================================="
print "explains the recording of additional information, for the rest see 1dnormal"
model = stats.norm.pdf
def testfunction():
return 3
startvalue = 0
proposal = sampler.Multivariate_Normal_Sampler(covariance=100)
mcmcsampler = mcmc.Metropolis_Like_MCMC(model, startvalue, proposal)
mcmcsampler.set_recording_function(testfunction, 10)
mcmcsampler.run(100)
mcmcsampler.chain.save_data()
print "done, look at the record file in the chain output folder"
print "here's a printout"
print mcmcsampler.chain.records
from scipy import stats
from matplotlib import pyplot
from fit import sampler, mcmc
def posterior(x):
return np.square(x)
# remember, np scales with exp( (x/scale)^2 /2)
def pdf(x):
return (stats.norm.pdf(x, scale = 1.0/np.sqrt(2) ))
startvalue = 0.2
proposal = sampler.Multivariate_Normal_Sampler(scale=2)
mcmcsampler = mcmc.Metropolis_Like_MCMC(posterior,startvalue, proposal, nll=True)
mcmcsampler.save_intervals = 1000
mcmcsampler.run(10000)
mcmcsampler.chain.convert_list_to_nparray()
#plotting
pyplot.figure(num=3)
pyplot.plot(np.arange(-1,1,0.01), posterior(np.arange(-1,1,0.01)))
pyplot.title("target distribution")
pyplot.hist(mcmcsampler.chain.statearray[:,0], bins=100, normed=True)
pyplot.plot(np.arange(-5,5,0.01), pdf(np.arange(-5,5,0.01)))
pyplot.show()
server.submit(target, (par, ), modules=("math", )) for par in x
]
out = [job() for job in jobs]
return out
else:
return target(x)
prior = bayesian.Uniform_Prior(lower=0)
post = bayesian.Parallel_Posterior(likelihood, prior, nll=False)
startvalue = 0.1
proposal = sampler.Multivariate_Normal_Sampler(scale=1)
mcmcsampler = mcmc.Metropolis_Like_MCMC(post.get_value,
startvalue,
proposal,
parallel_cores=cores)
mcmcsampler.run(5000)
server.print_stats()
print "====================================="
print("acceptancerate =" + str(mcmcsampler.chain.get_acceptance_rate()))
print "The number of MCMC steps done is: ", len(mcmcsampler.chain.states)
print "As the parallel mode creates new model runs just in case other runs are "
print "rejected, the final number of mcmc steps depends stochastically on the acceptance rate"
#recalculates the model values, can be usefulf for stochastic models
#mcmcsampler.recalculate = True
#mcmcsampler.run(5000)