Error in Bayesian Gaussian Mixture Model Example in Colab

[Siva Rajesh]

Hi, 

I am trying to run the Bayesian Gaussian Mixture Model example in Colab. 

Question 1) 

In the sampling step "[mix_probs, loc, chol_precision], is_accepted = sample()" , I run into a warning and execution gets stuck there with the following message: "WARNING: Nested component "transform_diagonal" in composition "chain_of_transform_diagonal_of_fill_triangular" operates on inputs with increased degrees of freedom. This may result in an incorrect log_det_jacobian."

Question 2) 

In this example, Hamiltonian Monte Carlo sampling was used. Instead, I am trying to use randomwalk metropolis with a Normal proposal distribution. 

This is the modified sampling function I use:

#################################

mh_normal_unconstrained = tfp.mcmc.TransformedTransitionKernel(

tfp.mcmc.RandomWalkMetropolis(

target_log_prob_fn=unnormalized_posterior_log_prob),

bijector = unconstraining_bijectors)

@tf.function(autograph=False)

def sample():

return tfp.mcmc.sample_chain(

num_results=2000,

num_burnin_steps=500,

current_state=initial_state,

kernel=mh_normal_unconstrained,

trace_fn=lambda _, kernel_results: kernel_results)

#################################

However, the results are pretty off from the true values. Can you please check if this is the correct code or any corrections are required. 

Thanks in advance,

Siva