Could you help me with dirichlet process mixture model？

[wangjiusi]

I am very sorry, I will reply you so late.
I use pm.DensityDist and the dimension of the model is reduced to one-dimensional data, then the experiment is re-run.
But there was an error.
Here is my code:

K = 8
n_samples = N
from pymc3.math import logsumexp
def KMM_logp(weight, aj, bj):
    def logp(value):
        Ncomps = 8
        logps = []
        for i in range(Ncomps):
            a = aj[i]*bj[i]
            b = aj[i]*(1 - bj[i])
            result = tt.gammaln(a + b) - tt.gammaln(a) - tt.gammaln(b) + (a - 1)*tt.log(value) + (b - 1)*tt.log(1 - value)
            logps.append(tt.log(weight[i]) + result)
        return tt.sum(logsumexp(tt.stacklists(logps)[:, :n_samples], axis=0))
    return logp

with pm.Model() as model:
    alpha = pm.Gamma('alpha', 1., 1.)
    beta = pm.Beta('beta', 1., alpha, shape=K)
    w = pm.Deterministic('w', stick_breaking(beta))
    
    Phi = pm.InverseGamma('Phi', alpha=1, beta=1)
    Delte = pm.Exponential('Delte', lam=1/8)
    Mu = pm.InverseGamma('Mu', alpha=1, beta=1)
    Eta = pm.Uniform('Eta', lower=0.0, upper=1.0)
    aj = pm.InverseGamma('aj', alpha=Phi, beta=Delte, shape=K)
    bj = pm.Beta('bj', alpha=Mu*Eta, beta=Mu*(1-Eta), shape=K)
    
    obs = pm.DensityDist('obs', KMM_logp(w, aj, bj), observed=X)

I use MCMC to inference, the error is:

AttributeError: Can't pickle local object 'KMM_logp.<locals>.logp'

while I use variational inference, no error, but has a bad result.