BUG: Sampling fails because 'numpy.random._generator.Generator' object has no attribute 'spawn'

[cluhmann]

Describe the issue:

PyMC now relies on the spawn() method from numpy.random._generator.Generator. But this is a relatively recent addition to numpy (added in v1.25). So we need to set a lower limit on the numpy dependency.

Reproduceable code example:

import pymc as pm

with pm.Model() as model:
    a = pm.Normal("a")
    obs = pm.Normal("obs", a, 1, observed=[1,2])
    idata = pm.sample()

Error message:

```shell --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) Input In [7], in () 7 a = pm.Normal("a") 8 obs = pm.Normal("obs", a, 1, observed=[1,2]) ----> 9 idata = pm.sample() 11 print(az.summary(idata)) File ~/mambaforge/envs/pymc/lib/python3.10/site-packages/pymc/sampling/mcmc.py:733, in sample(draws, tune, chains, cores, random _seed, progressbar, progressbar_theme, step, var_names, nuts_sampler, initvals, init, jitter_max_retries, n_init, trace, discard _tuned_samples, compute_convergence_checks, keep_warning_stat, return_inferencedata, idata_kwargs, nuts_sampler_kwargs, callback , mp_ctx, blas_cores, model, compile_kwargs, **kwargs) 731 if random_seed == -1: 732 random_seed = None --> 733 rngs = get_random_generator(random_seed).spawn(chains) 734 random_seed_list = [rng.integers(2**30) for rng in rngs] 736 if not discard_tuned_samples and not return_inferencedata: AttributeError: 'numpy.random._generator.Generator' object has no attribute 'spawn' ```

PyMC version information:

# packages in environment at /home/xian/mambaforge/envs/pymc: # # Name Version Build Channel numpy 1.24.2 py310h8deb116_0 conda-forge pymc 5.19.0 hd8ed1ab_0 conda-forge pymc-base 5.19.0 pyhd8ed1ab_0 conda-forge pytensor 2.26.4 py310ha549d7f_0 conda-forge pytensor-base 2.26.4 py310h89e8f5a_0 conda-forge

[nataziel]

This was true as of 5.18, I ran into the problem last week and didn't think to file a report sorry! From this commit: 4ea1406