def write_parametric(f, Bpar, runtime):
outs = "%g %g %g " % (Bpar.estimate[0],Bpar.getCI(1,(.025,)),Bpar.getCI(1,(.975))) # m.par.e m.par.l m.par.h
outs += "%g %g %g " % (Bpar.estimate[1],ptile(Bpar.mcestimates[:,1],2.5),ptile(Bpar.mcestimates[:,1],97.5)) # w.par.e w.par.l w.par.h
outs += "%g %g %g " % (Bpar.deviance, ptile(Bpar.mcdeviance,95), psigcorrect.estimate_nu (Bpar)[0]) # d.par d.par.crit nu.par
outs += "%g %g %g " % (Bpar.Rpd,ptile(Bpar.mcRpd,2.5),ptile(Bpar.mcRpd,97.5)) # rpd.par rpd.par.l rpd.par.h
outs += "%g %g %g " % (Bpar.Rkd,ptile(Bpar.mcRkd,2.5),ptile(Bpar.mcRkd,97.5)) # rkd.par rkd.par.l rkd.par.h
outs += ("%g "*options.nblocks) % tuple(Bpar.infl)
outs += "%g " % runtime
f.write(outs)
def write_parametric(f, Bpar, runtime):
outs = "%g %g %g " % (Bpar.estimate[0], Bpar.getCI(
1, (.025, )), Bpar.getCI(1, (.975))) # m.par.e m.par.l m.par.h
outs += "%g %g %g " % (Bpar.estimate[1], ptile(
Bpar.mcestimates[:, 1], 2.5), ptile(Bpar.mcestimates[:, 1], 97.5)
) # w.par.e w.par.l w.par.h
outs += "%g %g %g " % (Bpar.deviance, ptile(
Bpar.mcdeviance, 95), psigcorrect.estimate_nu(Bpar)[0]
) # d.par d.par.crit nu.par
outs += "%g %g %g " % (Bpar.Rpd, ptile(Bpar.mcRpd,
2.5), ptile(Bpar.mcRpd, 97.5)
) # rpd.par rpd.par.l rpd.par.h
outs += "%g %g %g " % (Bpar.Rkd, ptile(Bpar.mcRkd,
2.5), ptile(Bpar.mcRkd, 97.5)
) # rkd.par rkd.par.l rkd.par.h
outs += ("%g " * options.nblocks) % tuple(Bpar.infl)
outs += "%g " % runtime
f.write(outs)
def write_bayes(f, mcmc, mcmc_conv, runtime):
outs = "%g %g %g %g %g " % (
mcmc.estimate[0], # m.bay.m
mcmc.mapestimate[0], # m.bay.map
mcmc.posterior_median[0],# m.bay.median
mcmc.getCI(1,(.025,)), # m.bay.l
mcmc.getCI(1,(.975))) # m.bay.h
outs += "%g %g %g %g %g " % (
mcmc.estimate[1], # w.bay.m
mcmc.mapestimate[1], # w.bay.map
mcmc.posterior_median[1], # w.bay.median
ptile(mcmc.mcestimates[:,1],2.5), # w.bay.l
ptile(mcmc.mcestimates[:,1],97.5)) # w.bay.h
outs += "%g %g %g " % (mcmc.deviance,mcmc.bayesian_p('deviance'), psigcorrect.estimate_nu (mcmc)[0]) # d.bay d.bay.p
outs += "%g %g " % (mcmc.Rpd,mcmc.bayesian_p('Rpd')) # d.rpd d.rpd.p
outs += "%g %g " % (mcmc.Rkd,mcmc.bayesian_p('Rkd')) # d.rkd d.rkd.p
outs += "%d %g %g %g " % (mcmc_conv,mcmc.Rhat(0),mcmc.Rhat(1),mcmc.Rhat(2))
outs += ("%g "*options.nblocks) % tuple(mcmc.infl)
outs += "%g " % runtime
f.write(outs)
def write_bayes(f, mcmc, mcmc_conv, runtime):
outs = "%g %g %g %g %g " % (
mcmc.estimate[0], # m.bay.m
mcmc.mapestimate[0], # m.bay.map
mcmc.posterior_median[0], # m.bay.median
mcmc.getCI(1, (.025, )), # m.bay.l
mcmc.getCI(1, (.975))) # m.bay.h
outs += "%g %g %g %g %g " % (
mcmc.estimate[1], # w.bay.m
mcmc.mapestimate[1], # w.bay.map
mcmc.posterior_median[1], # w.bay.median
ptile(mcmc.mcestimates[:, 1], 2.5), # w.bay.l
ptile(mcmc.mcestimates[:, 1], 97.5)) # w.bay.h
outs += "%g %g %g " % (mcmc.deviance, mcmc.bayesian_p('deviance'),
psigcorrect.estimate_nu(mcmc)[0]) # d.bay d.bay.p
outs += "%g %g " % (mcmc.Rpd, mcmc.bayesian_p('Rpd')) # d.rpd d.rpd.p
outs += "%g %g " % (mcmc.Rkd, mcmc.bayesian_p('Rkd')) # d.rkd d.rkd.p
outs += "%d %g %g %g " % (mcmc_conv, mcmc.Rhat(0), mcmc.Rhat(1),
mcmc.Rhat(2))
outs += ("%g " * options.nblocks) % tuple(mcmc.infl)
outs += "%g " % runtime
f.write(outs)
print count_bay, mcmc.getCI(1,(.025,.975))
toc = time.time()
write_bayes(outfile, mcmc, mcmc_conv, toc-tic)
print "Done Bayes"
write_data(outfile, data)
outfile.write ("\n")
outfile.flush()
# datareduce section is not up-to-date
# current state is that it doesn't seem to help much for improving the results
# in case you want to use this you must first make it work again
if False:
data = array ( data )
dataml = data.copy ()
nu = psigcorrect.estimate_nu (Bpar)[0]
print "==============", nu, "==============="
dataml[:,1] = ( data[:,1] * nu ).astype("i")
dataml[:,2] = ( data[:,2] * nu ).astype("i")
print dataml
try:
Bnpr = pypsignifit.BootstrapInference ( dataml, sample=options.nbootstrap, priors=constraints, parametric=False, **ana_kwargs )
except:
sys.stderr ( "An error ocurred in simulation %d during nonparametric bootstrap\n" % ( simulation,) )
print "Done npar"
try:
Bpar = pypsignifit.BootstrapInference ( dataml, sample=options.nbootstrap, priors=constraints, parametric=True, **ana_kwargs )
except:
sys.stderr ( "An error ocurred in simulation %d during parametric bootstrap\n" % ( simulation,) )
print "Done par"
print count_bay, mcmc.getCI(1, (.025, .975))
toc = time.time()
write_bayes(outfile, mcmc, mcmc_conv, toc - tic)
print "Done Bayes"
write_data(outfile, data)
outfile.write("\n")
outfile.flush()
# datareduce section is not up-to-date
# current state is that it doesn't seem to help much for improving the results
# in case you want to use this you must first make it work again
if False:
data = array(data)
dataml = data.copy()
nu = psigcorrect.estimate_nu(Bpar)[0]
print "==============", nu, "==============="
dataml[:, 1] = (data[:, 1] * nu).astype("i")
dataml[:, 2] = (data[:, 2] * nu).astype("i")
print dataml
try:
Bnpr = pypsignifit.BootstrapInference(dataml,
sample=options.nbootstrap,
priors=constraints,
parametric=False,
**ana_kwargs)
except:
sys.stderr(
"An error ocurred in simulation %d during nonparametric bootstrap\n"
% (simulation, ))
print "Done npar"
