import pymc as pm
from numpy import *
from age_pr_datasets import *
N_pops = len(datasets)
S_guess = exp(-a/20.)
S_guess /= sum(S_guess)
# Model for central tendency of age distribution.
alph_pri = 10.
scg = []
for i in xrange(len(a)):
asg_now = pm.lam_dtrm('asg_%i'%i, lambda alpha=alph_pri, S=S_guess, i=i : alpha*S[i])
scg.append(pm.Gamma('scg_%i'%i, alpha=asg_now, beta=1., value=asg_now.value))
sc = pm.lam_dtrm('sc',lambda gams=scg: array(gams)/sum(gams))
# Model for age distributions of individual populations
# alph = pm.Gamma('alph',2.,2./1000.)
alph = pm.OneOverX('alph', value=50.)
asc = pm.lam_dtrm('asc',lambda alph=alph,S=sc: alph*S)
S = []
S_steps = []
data_list = []
for name in datasets.iterkeys():
this_dataset = datasets[name]
data_list = []
pred_data_list = []
deviance_list = []
pred_deviance_list = []
p_vec_list = []
P_prime_list = []
for name in datasets.iterkeys():
safe_name = name.replace('.','_')
P_prime_now = pm.Beta('P_prime_%s'%safe_name,3.,3.)
p_vec_now = pm.MvNormalChol('p_vec_%s'%safe_name, p_mean, cholfac)
p_vec_list.append(p_vec_now)
P_prime_list.append(P_prime_now)
b = pm.lam_dtrm('b', lambda p_vec = p_vec_now: 1./exp(p_vec[0]))
if methods[name] == 'Microscopy':
# alpha, s and c depend on p_vec[1:4]
c = pm.lam_dtrm('c', lambda p_vec = p_vec_now: 1./exp(p_vec[1]))
alph = pm.lam_dtrm('alph', lambda p_vec = p_vec_now: exp(p_vec[2]))
s = pm.lam_dtrm('s', lambda p_vec = p_vec_now: pm.invlogit(p_vec[3]))
elif methods[name] == 'RDT':
# alpha, s and c depend on p_vec[4:7]
c = pm.lam_dtrm('c', lambda p_vec = p_vec_now: 1./exp(p_vec[4]))
alph = pm.lam_dtrm('alph', lambda p_vec = p_vec_now: exp(p_vec[5]))
s = pm.lam_dtrm('s', lambda p_vec = p_vec_now: pm.invlogit(p_vec[6]))