Mixture of Gamma Distribution: Cannot find valid initial parameters. Out of Support Values

1

Hello everyone! I am very new to numpyro and despite trying to search online or on this forum for solutions, I am unable to reason out how to fix my code.

@config_enumerate
def gamma_mixture(data):

    K = 3
    # Global variables.
    weights = numpyro.sample("weights", dist.Dirichlet(jnp.ones(K)))
    with numpyro.plate("components", K):
        alpha = abs(numpyro.sample("alpha", dist.Uniform(0, 10)))
        beta = 1/abs(numpyro.sample("beta", dist.Uniform(0, 3)))

    with numpyro.plate("data", len(data)):
        # Local variables.
        assignment = numpyro.sample("assignment", dist.Categorical(weights))
        numpyro.sample("obs", dist.Gamma(concentration=alpha[assignment], rate=beta[assignment]), obs=data)

# Let's try to run MCMC
chosen_model = gamma_mixture
nuts_kernel = NUTS(model=chosen_model, target_accept_prob=0.80)
mcmc = MCMC(nuts_kernel, num_warmup=1000, num_samples=2000, num_chains=1)
mcmc.run(random.PRNGKey(0), data)
mcmc.print_summary()

I appreciate in advance for your assistance!

2

What values are in the array data?
i ask because the Gamma distribution is defined for all positive numbers.

If there are values that are zero or negative in data then this would be considered out of the support for the Gamma dist.

3

Thank you Andrew. I actually solved it, and forgot to close the thread. And yes, you are correct. I did not check to ensure the data must be in positive range.