def MCMC(whichx, nsamp, fname, nd, bigdata, burnin=500, run=500):
rho_pars = [-2., 6., .0065]
logg_pars = [-1.850, 7., .0065]
pars_init = logg_pars
if whichx == "rho":
pars_init = rho_pars
x, y, xerr, yerr = load_data(whichx, nd=nd, bigdata=True)
# format data and generate samples
obs = np.vstack((x, y))
u = np.vstack((xerr, yerr))
up = np.vstack((xerr, yerr))
um = np.vstack((xerr*.5, yerr*.5))
# s = generate_samples_log(obs, up, um, nsamp) # FIXME
s = generate_samples(obs, u, nsamp) # FIXME
# if nsamp == 1:
# s[0, :, :] = x
# s[1, :, :] = y
# print np.shape(s)
# assert 0
# set up and run emcee
ndim, nwalkers = len(pars_init), 32
pos = [pars_init + 1e-4*np.random.randn(ndim) for i in range(nwalkers)]
sampler = emcee.EnsembleSampler(nwalkers, ndim, lnprob,
args=(s, obs, u))
print "burning in..."
pos, _, _, = sampler.run_mcmc(pos, burnin)
sampler.reset()
print "production run..."
sampler.run_mcmc(pos, run)
samp = sampler.chain[:, 50:, :].reshape((-1, ndim))
m, c, sig = map(lambda v: (v[1], v[2]-v[1], v[1]-v[0]),
zip(*np.percentile(samp, [16, 50, 84], axis=0)))
pars = [m[0], c[0], sig[0]]
# save samples
f = h5py.File("%s_samples_%s.h5" % (whichx, fname), "w")
data = f.create_dataset("samples", np.shape(samp))
data[:, 0] = samp[:, 0]
data[:, 1] = samp[:, 1]
data[:, 2] = samp[:, 2]
f.close()
def MCMC(whichx, nsamp, fname, nd, bigdata, burnin=500, run=500):
rho_pars = [-2., 6., .0065]
logg_pars = [-1.850, 7., .0065]
pars_init = logg_pars
if whichx == "rho":
pars_init = rho_pars
x, y, xerr, yerr = load_data(whichx, nd=nd, bigdata=True)
# format data and generate samples
obs = np.vstack((x, y))
u = np.vstack((xerr, yerr))
up = np.vstack((xerr, yerr))
um = np.vstack((xerr * .5, yerr * .5))
# s = generate_samples_log(obs, up, um, nsamp) # FIXME
s = generate_samples(obs, u, nsamp) # FIXME
# if nsamp == 1:
# s[0, :, :] = x
# s[1, :, :] = y
# print np.shape(s)
# assert 0
# set up and run emcee
ndim, nwalkers = len(pars_init), 32
pos = [pars_init + 1e-4 * np.random.randn(ndim) for i in range(nwalkers)]
sampler = emcee.EnsembleSampler(nwalkers, ndim, lnprob, args=(s, obs, u))
print "burning in..."
pos, _, _, = sampler.run_mcmc(pos, burnin)
sampler.reset()
print "production run..."
sampler.run_mcmc(pos, run)
samp = sampler.chain[:, 50:, :].reshape((-1, ndim))
m, c, sig = map(lambda v: (v[1], v[2] - v[1], v[1] - v[0]),
zip(*np.percentile(samp, [16, 50, 84], axis=0)))
pars = [m[0], c[0], sig[0]]
# save samples
f = h5py.File("%s_samples_%s.h5" % (whichx, fname), "w")
data = f.create_dataset("samples", np.shape(samp))
data[:, 0] = samp[:, 0]
data[:, 1] = samp[:, 1]
data[:, 2] = samp[:, 2]
f.close()
def lnprob(pars, samples, obs, u):
return lnlikeH(pars, samples, obs, u) + lnprior(pars)
if __name__ == "__main__":
# load data
# x = f, y = rho, z = teff
x, xerr, y, yerr, z, zerr = np.genfromtxt("data/log.dat").T
obs = np.vstack((x, y))
u = np.vstack((xerr, yerr))
nsamp = 3
s = generate_samples(obs, u, nsamp)
pars_init = [-1.850, 5.413, .065]
ndim, nwalkers = len(pars_init), 32
pos = [pars_init + 1e-4 * np.random.randn(ndim) for i in range(nwalkers)]
sampler = emcee.EnsembleSampler(nwalkers, ndim, lnprob, args=(s, obs, u))
print "burning in..."
pos, _, _, = sampler.run_mcmc(pos, 500)
sampler.reset()
print "production run..."
sampler.run_mcmc(pos, 1000)
samp = sampler.chain[:, 50:, :].reshape((-1, ndim))
m, c, lnf = map(lambda v: (v[1], v[2] - v[1], v[1] - v[0]),
zip(*np.percentile(samp, [16, 50, 84], axis=0)))
def lnprior(pars):
return 0.
def lnprob(pars, samples, obs, u):
return lnlikeH(pars, samples, obs, u) + lnprior(pars)
if __name__ == "__main__":
# load data
# x = f, y = rho, z = teff
x, xerr, y, yerr, z, zerr = np.genfromtxt("data/log.dat").T
obs = np.vstack((x, y))
u = np.vstack((xerr, yerr))
nsamp = 3
s = generate_samples(obs, u, nsamp)
pars_init = [-1.850, 5.413, .065]
ndim, nwalkers = len(pars_init), 32
pos = [pars_init + 1e-4*np.random.randn(ndim) for i in range(nwalkers)]
sampler = emcee.EnsembleSampler(nwalkers, ndim, lnprob,
args=(s, obs, u))
print "burning in..."
pos, _, _, = sampler.run_mcmc(pos, 500)
sampler.reset()
print "production run..."
sampler.run_mcmc(pos, 1000)
samp = sampler.chain[:, 50:, :].reshape((-1, ndim))
m, c, lnf = map(lambda v: (v[1], v[2]-v[1], v[1]-v[0]),
def MCMC(whichx, nsamp, fname, nd, extra, f, bigdata, burnin=500, run=1000):
"""
nsamp (int) = number of samples.
whichx (str) = logg or rho.
fname (str) = the name for saving all output
nd (int) = number of data points (for truncation).
If this is zero, all the data are used.
bigdata (boolean) which data file to use.
"""
# set initial parameters
if extra:
rho_pars = [-1.793214679, 5.34215688, 0.02334097, .0002600777]
logg_pars = [-1.02143776, 5.69156647, .24239756, .049233887]
else:
rho_pars = [-1.69293833, 5.1408906, .0065]
logg_pars = [-1.05043614, 5.66819525, .0065]
pars_init = logg_pars
if whichx == "rho":
pars_init = rho_pars
# load the data
x, y, xerr, yerr = load_data(whichx, nd=nd, bigdata=False)
# format data and generate samples
obs = np.vstack((x, y))
u = np.vstack((xerr, yerr))
up = np.vstack((xerr, yerr))
um = np.vstack((xerr*.5, yerr*.5))
s = generate_samples(obs, u, nsamp)
# set up and run emcee
ndim, nwalkers = len(pars_init), 32
pos = [pars_init + 1e-4*np.random.randn(ndim) for i in range(nwalkers)]
sampler = emcee.EnsembleSampler(nwalkers, ndim, lnprob,
args=(s, obs, u, extra, f))
print "burning in..."
pos, _, _, _ = sampler.run_mcmc(pos, burnin)
sampler.reset()
print "production run..."
sampler.run_mcmc(pos, run)
# load likelihood
lls = sampler.blobs
flat_lls = np.reshape(lls, (np.shape(lls)[0]*np.shape(lls)[1]))
samp = np.vstack((sampler.chain[:, :, :].reshape(-1, ndim).T, flat_lls)).T
sa = samp.T[0]
print type(sa)
print np.isfinite(sa)
print sa
print np.shape(sa), np.shape(sa[np.isfinite(sa)])
# save samples
f = h5py.File("%s_samples_%s.h5" % (whichx, fname), "w")
data = f.create_dataset("samples", np.shape(samp))
data[:, 0] = samp[:, 0]
data[:, 1] = samp[:, 1]
data[:, 2] = samp[:, 2]
data[:, 3] = samp[:, 3]
print "samp shape = ", np.shape(samp), np.shape(data)
if extra:
data[:, 4] = samp[:, 4]
f.close()
def MCMC(whichx, nsamp, fname, nd, extra, f, bigdata, burnin=500, run=1000):
"""
nsamp (int) = number of samples.
whichx (str) = logg or rho.
fname (str) = the name for saving all output
nd (int) = number of data points (for truncation).
If this is zero, all the data are used.
bigdata (boolean) which data file to use.
"""
# set initial parameters
if extra:
rho_pars = [-1.793214679, 5.34215688, 0.02334097, .0002600777]
logg_pars = [-1.02143776, 5.69156647, .24239756, .049233887]
else:
rho_pars = [-1.69293833, 5.1408906, .0065]
logg_pars = [-1.05043614, 5.66819525, .0065]
pars_init = logg_pars
if whichx == "rho":
pars_init = rho_pars
# load the data
x, y, xerr, yerr = load_data(whichx, nd=nd, bigdata=False)
# format data and generate samples
obs = np.vstack((x, y))
u = np.vstack((xerr, yerr))
up = np.vstack((xerr, yerr))
um = np.vstack((xerr * .5, yerr * .5))
s = generate_samples(obs, u, nsamp)
# set up and run emcee
ndim, nwalkers = len(pars_init), 32
pos = [pars_init + 1e-4 * np.random.randn(ndim) for i in range(nwalkers)]
sampler = emcee.EnsembleSampler(nwalkers,
ndim,
lnprob,
args=(s, obs, u, extra, f))
print "burning in..."
pos, _, _, _ = sampler.run_mcmc(pos, burnin)
sampler.reset()
print "production run..."
sampler.run_mcmc(pos, run)
# load likelihood
lls = sampler.blobs
flat_lls = np.reshape(lls, (np.shape(lls)[0] * np.shape(lls)[1]))
samp = np.vstack((sampler.chain[:, :, :].reshape(-1, ndim).T, flat_lls)).T
sa = samp.T[0]
print type(sa)
print np.isfinite(sa)
print sa
print np.shape(sa), np.shape(sa[np.isfinite(sa)])
# save samples
f = h5py.File("%s_samples_%s.h5" % (whichx, fname), "w")
data = f.create_dataset("samples", np.shape(samp))
data[:, 0] = samp[:, 0]
data[:, 1] = samp[:, 1]
data[:, 2] = samp[:, 2]
data[:, 3] = samp[:, 3]
print "samp shape = ", np.shape(samp), np.shape(data)
if extra:
data[:, 4] = samp[:, 4]
f.close()
