Try to sample from a simple custom-defined uniform distribution but got warnings

1 
import numpy as np
import pymc as pm
from pytensor.tensor import TensorVariable
import matplotlib.pyplot as plt
import arviz as az
import pytensor.tensor as pt
from pymc.sampling.jax import sample_blackjax_nuts, sample_numpyro_nuts
def logp(value, lower, upper):
    res = pt.switch(
        pt.bitwise_and(pt.ge(value, lower), pt.le(value, upper)),
        pt.fill(value, -pt.log(upper - lower)),
        -np.inf,
    )
    return res
with pm.Model():
    lower = 0
    upper = 10
    pm.CustomDist(
        'a',
        lower,
        upper,
        logp=logp,
    )
    idata = sample_numpyro_nuts(draws=1000, tune=1000, chains=4, target_accept=0.96, random_seed=123,
                                      chain_method='sequential', compute_convergence_checks = True,                                      progressbar = True,postprocessing_backend = 'gpu')

Got warning: There were 3907 divergences after tuning. Increase target_accept or reparameterize.
The rhat statistic is larger than 1.01 for some parameters. This indicates problems during sampling.

using az.plot_trace() shows that it is uniform distributed.

2

You probably need to pass an interval transform so the sampler can’t propose values with zero probability. That’s what Uniform would do by default

3

Hi. Thank you. Do you mean like this? it didn’t help.

    interval = pm.distributions.transforms.Interval(lower=0, upper=10)
    lower = 0
    upper = 10
    pm.CustomDist(
        'a',
        lower,
        upper,
        logp=logp,
        transform=interval,
    )