MCMC Sampler¶
The MCMC sampling is done using emcee based CosmoHammer.
For more details about
emcee: https://emcee.readthedocs.io/en/stable/
CosmoHammer: http://cosmo-docs.phys.ethz.ch/cosmoHammer/
[1]:
import emcee
import cosmoHammer
import numpy as np
we have used an older version of emcee.
[2]:
emcee.__version__,cosmoHammer.__version__
[2]:
(u'2.2.1', '0.6.1')
All the custom core and likelihood modules are already in the EmuPBk’s MCMC class.
[3]:
from EmuPBk.MCMC import sampler
Loading the data
[4]:
path = '/home/ht/PycharmProjects/EmuPBK/vv/data/data_Powerspectrum/'
data = np.loadtxt(path+'pk_test')
nbins = np.loadtxt(path+'nbins_test')
[5]:
mcmc = sampler.Run_MCMC(data[0],nbins[0])
'''
:param data: load your data
:param cov: data for covariance calculation
'''
[5]:
'\n:param data: load your data\n:param cov: data for covariance calculation\n'
MCMC using already existing ANN models¶
[7]:
mcmc.load_existing_model(name = 'Pk')
'''
Use the existing ANN models for MCMC analysis
:param name: use ('Pk','Bk02','Bk03','Bk15')==>for powerspectrum, Bispectrum02, Bispectrum03, Bispectrum15
'''
Core setup is done
Core setup is done
logLikelihood setup is done
[9]:
mcmc.sampler(walker_ratio=6, burnin=200, samples=200, threads=-1)
find best fit point
converged after 131 iterations!
best fit found: [[-1.32994776e-05]] [ 16.01032703 31.67938218 526.2873946 ]
start sampling:.
The time taken 29.50 sec. done!
Done!
MCMC using Your own model¶
[ ]:
mcmc.load_model(load_model,name,rescale)
'''
:param load_model: load your own model, (give the path)
:param name: name of data, ('Pk','Bk02','Bk03','Bk15')==>for powerspectrum, Bispectrum02, Bispectrum03, Bispectrum15
:param rescale: rescale used in the training
'''
mcmc.sampler(walker_ratio=6, burnin=200, samples=200, threads=-1)