0

I am doing a Bayesian analysis with pymc3 (a graphic model with a lot of priors) as below:

import pymc3 as pm
import theano.tensor as T
with pm.Model() as smk:
    a_u = pm.InverseGamma('a_u',alpha=2*np.ones(L),beta=1/A_u**2,shape=L)
    sigma2_u = 
    pm.InverseGamma('sigma2_u',alpha=2*np.ones(L),beta=1/a_u,shape=L)
    a_eps = pm.InverseGamma('a_eps',alpha=2,beta=1/A_eps**2)
    sigma2_eps = pm.InverseGamma('sigma2_eps',alpha=2,beta=1/a_eps)
    a_R = pm.InverseGamma('a_R',alpha=2,beta=1/A_R**2,shape=q)
    Sigma_R_inv = pm.distributions.multivariate.Wishart('Sigma_R_inv',nu=v+q-1,V=1/(2*v)*T.nlinalg.AllocDiag(a_R))
    Sigma_R = T.nlinalg.MatrixInverse(Sigma_R_inv)
    u_R = pm.distributions.multivariate.MvNormal('u_R',np.zeros(m*q),T.slinalg.kron(np.eye(m),Sigma_R))
    u_G = pm.distributions.multivariate.MvNormal('u_G',np.zeros(L),T.nlinalg.AllocDiag(sigma2_u))
    beta = pm.Normal('beta',mu=0,sd=sigma_beta,shape=p)
    y = pm.Normal('y',mu = T.dot(X,beta)+T.dot(Z_R,u_R)+T.dot(Z_G,u_G),sd=sigma_eps,observed=Y) 
with smk:
    trace= pm.sample(10000)

Then it reports an error, "TypeError: object() takes no parameters" on this line,

Sigma_R_inv = pm.distributions.multivariate.Wishart('Sigma_R_inv',nu=v+q-1,V=1/(2*v)*T.nlinalg.AllocDiag(a_R))

I am wondering why, could someone help me out, thanks!

Greyson
  • 1
  • 1
    Looks like this might be the problem: https://stackoverflow.com/questions/30692742/creating-square-matrix-with-eigenvalues-on-the-diagonal-in-theano – colcarroll Apr 06 '18 at 14:40
  • 1
    (also note all the warnings about not using `Wishart` in PyMC3!) – colcarroll Apr 06 '18 at 14:40

0 Answers0