def runToy2D():
np.random.seed(0)
random.seed(0)
for i in range(20):
print ' '*random.randint(0,70) + '*'
# priors:
aP = prior.uniform(-5,5) # will have 2D gaussian toy likelihood
bP = prior.uniform(-5,5)
g.priors = [aP,bP]
# parameters ('name', x_init, isFrozen):
a = particle.param('a',aP.sample(),False)
b = particle.param('b',bP.sample(),False)
g.initParams = [a,b]
for i,p in enumerate(g.initParams): # important!
p.setPrior(i)
# which indexes are thawed:
for i,p in enumerate(g.initParams):
if(not p.isFrozen):
g.thawedIdxs.append(i)
# choose likelihood:
g.likelihood = likelihood.toyGauss()
# mass vector for parameters:
g.masses = np.ones(len(g.initParams))
# create the initial particle set:
particles = []
for i in range(g.nParticles):
particles.append(particle.particle(g.initParams))
particles[i].assignPriorSample()
# set the inference running:
fname = '/Users/jmrv/Documents/school/mit/research/software/nest/samples/toy2d.txt'
nested.sample(particles,fname)
pos = posterior.posterior(fname)
plotname = '/Users/jmrv/Documents/school/mit/research/software/nest/plots/toy2d.pdf'
pos.plotMarginals(plotname)
def runOneChain(iChain,convConn):
# priors: (min, max, logarithmic?)
lambda0p = prior.uniform(1e-3,1e2,True)
dldtp = prior.uniform(-2,2,False)
g.priors = [lambda0p,dldtp]
# initial parameters: (prior, starting value, frozen?)
lambda0 = cob.param('lambda0',5,False) # poisson rate at start of dataset
lambda0.setPrior(0)
dldt = cob.param('dldt',0.2,False) # rate of change of poisson rate
dldt.setPrior(1)
g.initParams = [lambda0,dldt]
# which indexes are thawed:
for i,p in enumerate(g.initParams):
if(not p.isFrozen):
g.thawedIdxs.append(i)
# mass vector for parameters:
g.masses = ones(len(g.initParams)) # (ensures double precision)
# data and likelihood:
counts = random.poisson(10,size=100)
time = range(len(counts))
data = transpose([time,counts])
g.likelihood = likelihood.poissDrift(data)
#g.likelihood = likelihood.multiGauss()
# start the chain:
fname = 'chains/chain-'+str(iChain)+'.txt'
ch = cob.chain(fname,convConn)
# evolve the chain:
# (the convergence monitor will only send anything to this end of
# the pipe once all chains have converged. evolve until then.)
while not convConn.poll():
ch.evolve()
def runToy1D():
np.random.seed(1)
random.seed(1)
for i in range(20):
print ' '*random.randint(0,70) + '*'
# priors:
aP = prior.uniform(-5,5) # single parameter. will have gaussian toy likelihood
g.priors = [aP]
# parameters ('name', x_init, isFrozen):
a = particle.param('a',aP.sample(),False)
g.initParams = [a]
for i,p in enumerate(g.initParams): # important!
p.setPrior(i)
# which indexes are thawed:
for i,p in enumerate(g.initParams):
if(not p.isFrozen):
g.thawedIdxs.append(i)
# choose likelihood:
g.likelihood = likelihood.toy1D()
# mass vector for parameters:
g.masses = np.ones(len(g.initParams))*0.05
# create the initial particle set:
particles = []
for i in range(g.nParticles):
particles.append(particle.particle(g.initParams,g.initStep))
particles[i].assignPriorSample()
# set the inference running:
nested.sample(particles)
def runOneChain(iChain,convConn):
# priors:
normp = prior.uniform(10**-4,10**-2.5,isLog=True) # powerlaw norm
alphap = prior.uniform(2.0,3.5) # powerlaw power
nHp = prior.uniform(0.5,3.0) # nH (absorption)
scareap = prior.uniform(1e-10,1e-6,isLog=True) # Sc line area
area1p = prior.uniform(10**-8,10**-5,isLog=True) # nuisance line 1
center1p = prior.uniform(3.5,3.75) #
sigma1p = prior.uniform(0.001,0.1) # natural width
area2p = prior.uniform(10**-7,10**-5,isLog=True) # nuisance line 2
center2p = prior.uniform(3.75,4.0) #
sigma2p = prior.uniform(0.001,0.1) # natural width
g.priors = [normp,alphap,nHp,scareap,area1p,center1p,sigma1p,area2p,center2p,sigma2p]
# parameters ('name', x_init, isFrozen):
names = ['norm', 'alpha', 'nH', 'Sc area', 'area1', 'center1', 'sigma1',
'area2', 'center2', 'sigma2']
g.initParams = []
for i in range(len(g.priors)):
if i == 2: # nH set by hwang '12
g.initParams.append(cob.param(names[i], 2.0, True))
else:
g.initParams.append(cob.param(names[i], g.priors[i].sample(), False))
for i,p in enumerate(g.initParams):
p.setPrior(i)
# which indices are thawed:
for i,p in enumerate(g.initParams):
if(not p.isFrozen):
g.thawedIdxs.append(i)
# mass vector for parameters:
g.masses = ones(len(g.initParams))
g.masses[0] = 1.0 # powerlaw norm
g.masses[1] = 1.0 # powerlaw alpha
#g.masses[2] = 1.0 # nH
g.masses[3] = 1.0 # Sc area
g.masses[4] = 1.0 # area 1
g.masses[5] = 1.0 # center 1
g.masses[6] = 0.5 # sigma 1
g.masses[7] = 1.0 # area 2
g.masses[8] = 1.0 # center 2
g.masses[9] = 0.5 # sigma 2
# data and likelihood:
epochs = ['4634','4635','4636','4637','4638']
#epochs = ['4637']
datasets = []
fnames = []
for e in epochs:
#froot = g.datadir+'cpoor'
froot = g.datadir+'eastegg'+'-'+e
fnames.append(froot)
datasets.append(cob.dataset(froot, [3.4,5],
g.datadir+'../bkgd', g.Aeastegg/g.Abkgd))
g.likelihood = likelihood.ScLike(datasets, g.codedir+'models/phabs1e22.txt')
g.dumpInfo(g.initParams, g.priors, g.masses, g.likelihood,fnames=fnames)
# start the chain:
fname = g.chaindir+'chain-'+str(iChain)+'.txt'
ch = cob.chain(fname,convConn)
# evolve the chain:
# (the convergence monitor will only send anything to this end of
# the pipe once all chains have converged. evolve until then.)
while not convConn.poll():
ch.evolve()
def runSc():
np.random.seed(0)
random.seed(0)
for i in range(20):
print ' '*random.randint(0,70) + '*'
# priors:
normP = prior.uniform(1e-2,1,isLog=True) # powerlaw norm
alphaP = prior.uniform(2,4) # powerlaw power
nHP = prior.uniform(1.0,3.0) # nH (absorption)
scAreaP = prior.uniform(1e-6, 1e-3, isLog=True) # Sc line area
area1P = prior.uniform(1e-6, 1e-3, isLog=True) # nuisance line 1
center1P = prior.uniform(3.5, 3.7) #
sigma1P = prior.uniform(0.001, 0.010) # natural width
area2P = prior.uniform(1e-5, 1e-4, isLog=True) # nuisance line 2
center2P = prior.uniform(3.75, 4.0) #
sigma2P = prior.uniform(0.001, 0.010) #
g.priors = [ normP, alphaP, \
nHP, \
scAreaP, \
#area1P,center1P,sigma1P, \
area2P,center2P,sigma2P ]
# parameters ('name', x_init, isFrozen):
norm = particle.param('norm',5e-2,False)
alpha = particle.param('alpha',2.9,False)
nH = particle.param('nH',2.0,True)
scarea = particle.param('Sc area',1e-5,False)
area1 = particle.param('area1',1e-5,False)
center1 = particle.param('center1',3.6,False)
sigma1 = particle.param('sigma1',0.005,False)
area2 = particle.param('area2',1.8e-5,False)
center2 = particle.param('center2',3.87,False)
sigma2 = particle.param('sigma2',0.005,False)
g.initParams = [ norm,alpha, \
nH, \
scarea, \
#area1,center1,sigma1, \
area2,center2,sigma2]
for i,p in enumerate(g.initParams):
p.setPrior(i)
# which indexes are thawed:
for i,p in enumerate(g.initParams):
if(not p.isFrozen):
g.thawedIdxs.append(i)
# mass vector for parameters:
g.masses = np.ones(len(g.initParams))*0.1
# data and likelihood:
g.likelihood = likelihood.ScLike( g.datadir+'column.warf',
g.datadir+'column.wrmf',
g.datadir+'column.pi',
g.modeldir+'phabs1e22.txt',
[3.4,5])
# create the initial particle set:
particles = []
for i in range(g.nParticles):
particles.append(particle.particle(g.initParams))
particles[i].assignPriorSample()
# set the inference running:
fname = g.sampleDir+'samples.txt'
nested.sample(particles,fname)
pos = posterior.posterior(fname)
plotname = g.plotDir+'sctest.pdf'
pos.plotMarginals(plotname)
# ---------------
# ~~~ todo and wishlist ~~~
# hey, likelihoods - separate class
# hey, plotting evolution to verify chmc
# hey, parallel processing and convergence
# hey, debugging gui
# ~~~~~~~~~~~~~~~~~~~~~~~~~
# chmc-specific modules:
import chainObject as cob
import prior
from posterior import *
# initial parameters:
kTp = prior.uniform(0.5,8.0,False) # (min, max, logarithmic?)
kT = cob.param('kT',kTp,1,False) # (prior, starting value, frozen?)
eltp = prior.uniform(1e-6,1000,True)
elt = cob.param('O',eltp,2,True) # 'elemental abundance'
taup = prior.uniform(1e9,1e12,True)
tau = cob.param('tau',taup,1.2e10,False)
initParams = [kT,elt,tau]
masses = ones(len(initParams))
# start the chain:
fp = open('chains/chain.txt','w')
ch = cob.chain(initParams,fp,masses) # (seedparams,file pointer, masses)
# evolve the chain:
for i in range(1000):
from numpy import * from matplotlib.pyplot import * # chmc-specific modules: from posterior import * import chainObject as cob import prior import chmcGlobals as g import likelihood # priors: normp = prior.uniform(1e-4,1,isLog=True) # powerlaw norm alphap = prior.uniform(1,5) # powerlaw power nhp = prior.uniform(1e-1,1e1,isLog=True) # nH (absorption) ~ tighten? dp = prior.normal(3.4,0.2) # distance to Cas A, kpc # reed 1995, apj 440, 706 agep = prior.uniform(2011-1620,2011-1700) # Cas A's age ~ weak beta timassp = prior.uniform(1e-8,1e0,isLog=True) # mass of 44Ti, in M_sol sccenterp = prior.normal(4.1,0.2) # center of sc line scsigmap = prior.uniform(0,1) # width of sc line ~ tighter, normal? area1p = prior.uniform(1e-8,1,isLog=True) # nuisance line 1 center1p = prior.uniform(3.55,0.2) # sigma1p = prior.uniform(0,1) # ~ tighter, normal? area2p = prior.uniform(1e-8,1,isLog=True) # nuisance line 2 center2p = prior.normal(3.75,0.2) # sigma2p = prior.uniform(0,1) # ~ tighter, normal? g.priors = [ normp,alphap,nhp, \ dp,agep,timassp, \ sccenterp,scsigmap, \ area1p,center1p,sigma1p, \ area2p,center2p,sigma2p]
def runOneChain(iChain,convConn):
# priors:
normp = prior.uniform(1e-6,1e1,isLog=True) # powerlaw norm
alphap = prior.uniform(1,5) # powerlaw power
nHp = prior.uniform(0.5,3.0) # nH (absorption)
# based on hwang+ 2012
scareap = prior.uniform(1e-9,1e-3,isLog=True) # Sc line area
area1p = prior.uniform(1e-9,1e-3,isLog=True) # nuisance line 1
center1p = prior.uniform(3.5,3.75) #
sigma1p = prior.uniform(0.001,0.1) # natural width
area2p = prior.uniform(1e-9,1e-2,isLog=True) # nuisance line 2
center2p = prior.uniform(3.75,4.0) #
sigma2p = prior.uniform(0.001,0.1) # natural width
g.priors = [ normp,alphap, \
nHp, \
scareap, \
area1p,center1p,sigma1p, \
area2p,center2p,sigma2p]
# parameters ('name', x_init, isFrozen):
norm = cob.param('norm',10**-1.4,False)
alpha = cob.param('alpha',3.3,False)
nH = cob.param('nH',2.0,True)
scarea = cob.param('Sc area',1e-7,False)
area1 = cob.param('area1',10**-5.2,False)
center1 = cob.param('center1',3.69,False)
sigma1 = cob.param('sigma1',0.005,False)
area2 = cob.param('area2',10**-4.5,False)
center2 = cob.param('center2',3.86,False)
sigma2 = cob.param('sigma2',0.01,False)
g.initParams = [ norm,alpha, \
nH, \
scarea, \
area1,center1,sigma1, \
area2,center2,sigma2]
for i,p in enumerate(g.initParams):
p.setPrior(i)
# which indexes are thawed:
for i,p in enumerate(g.initParams):
if(not p.isFrozen):
g.thawedIdxs.append(i)
# mass vector for parameters:
g.masses = ones(len(g.initParams))
g.masses[0] = 0.2 # powerlaw norm
g.masses[1] = 0.2 # powerlaw alpha
#g.masses[2] = 0.1 # nH
g.masses[3] = 0.01 # Sc area
g.masses[4] = 0.01 # area 1
g.masses[5] = 0.01 # center 1
g.masses[6] = 0.01 # sigma 1
g.masses[7] = 0.01 # area 2
g.masses[8] = 0.01 # center 2
g.masses[9] = 0.01 # sigma 2
# data and likelihood:
#epochs = ['4634','4635','4636','4637','4638']
epochs = ['4637']
datasets = []
fnames = []
for e in epochs:
#froot = g.datadir+'fe'
froot = g.datadir+'eastegg'+'-'+e
fnames.append(froot)
datasets.append(cob.dataset(froot, [3.4,5],
g.datadir+'../bkgd', g.Aeastegg/g.Abkgd))
g.likelihood = likelihood.ScLike(datasets, g.codedir+'models/phabs1e22.txt')
g.dumpInfo(g.initParams, g.priors, g.masses, g.likelihood,fnames=fnames)
# start the chain:
fname = g.chaindir+'chain-'+str(iChain)+'.txt'
ch = cob.chain(fname,convConn)
# evolve the chain:
# (the convergence monitor will only send anything to this end of
# the pipe once all chains have converged. evolve until then.)
while not convConn.poll():
ch.evolve()