Hierarchical Dirichlet Process in PyMC3

Question

I'm trying to implement Hierarchical Dirichlet Process (HDP) topic model using PyMC3. The HDP graphical model is shown below:

I came up with the following code:

import numpy as np
import scipy as sp
import pandas as pd

import seaborn as sns
import matplotlib.pyplot as plt

import pymc3 as pm
from theano import tensor as tt

np.random.seed(0)

def stick_breaking(beta):
    portion_remaining = tt.concatenate([[1], tt.extra_ops.cumprod(1 - beta)[:-1]])
    return beta * portion_remaining

def main():

    #load data    
    data = np.array([[1, 1, 1, 1], [1, 1, 1, 1], [0, 0, 0, 0]])    
    Wd = [len(doc) for doc in data]

    #HDP parameters    
    T = 10   # top-level truncation
    K = 2    # group-level truncation    
    V = 4    # number of words
    D = 3    # number of documents    

    with pm.Model() as model:    

        #top-level stick breaking
        gamma = pm.Gamma('gamma', 1., 1.)
        beta_prime = pm.Beta('beta_prime', 1., gamma, shape=T)
        beta = pm.Deterministic('beta', stick_breaking(beta_prime))

        #group-level stick breaking                                                
        alpha = pm.Gamma('alpha', 1., 1.)        
        pi_prime = pm.Beta("pi_prime", 1, alpha, shape=K) #Sethuraman's stick breaking
        #pi_prime = [pm.Beta("pi_prime_%s_%s" %(j,k), alpha*(beta[k]), alpha*(1-np.sum(beta[:k+1])), shape=1)
        #            for j in range(K) for k in range(T)]  #Teh's stick breaking
        pi = pm.Deterministic('pi', stick_breaking(pi_prime))

        #top-level DP
        H = pm.Dirichlet("H", a=np.ones(V), shape=V)
        phi_top = pm.Multinomial('phi_top', n=np.sum(Wd), p=H, shape=(T,V))                        
        G0 = pm.Mixture('G0', w=beta, comp_dists=phi_top)

        #group-level DP
        phi_group = [pm.Multinomial('phi_group_%s' %j, n=Wd[j], p=G0) for j in range(D)]
        Gj = [pm.Mixture('G_%s' %j, w=pi, comp_dists=phi_group[j]) for j in range(D)]

        #likelihood                
        w = [pm.Categorical("w_%s_%s" %(j,n), p = Gj[j], observed=data[j][n]) for j in range(D) for n in range(Wd[j])]      

    with model:
        trace = pm.sample(2000, n_init=1000, random_seed=42)


    pm.traceplot(trace)
    plt.show()


if __name__ == '__main__':
    main()

However, I'm currently getting an AssertionError that prevents me from debugging the rest of the model, it occurs at the following line:

phi_top = pm.Multinomial('phi_top', n=np.sum(Wd), p=H, shape=(T,V))

There's no additional information about the error. Does anyone know how to resolve this?

Hierarchical Dirichlet Process in PyMC3

0 Answers