Hi all,
I’ve built a HLM and have been able to run the model with the numpyro
sampler and the nuts
sampler. The results of the model look promising and the fits are good. I am trying to speed it up even further though, and came across the nutpie
sampler.
I have followed the docs here, but I fail on this line:
compiled_model = nutpie.compile_pymc_model(pymc_model)
with the error message:
TypingError: Failed in nopython mode pipeline (step: inline inlinable functions)
Failed in untyped mode pipeline (step: handles literal_unroll)
Failed in literal_unroll_subpipeline mode pipeline (step: performs partial type inference)
Untyped global name 'tuple_setitem_literal': Cannot determine Numba type of <class 'function'>
File "path/to/nutpie/compile_pymc.py", line 497:
def extract_shared(x, user_data_):
<source elided>
dat = extract_array(user_data["shared"], index)
_shared_tuple = tuple_setitem_literal(_shared_tuple, index, dat)
^
Is this something anyone else has come across? I am wondering if I need to redefine my model in some way, or if it is a bug given that I am able to run it with other samplers? The same error occurs if I call nuts_sampler = "nutpie"
inside of pm.sample()
.
Also, are there plans to be able to pass the var_names
argument for nutpie
?
Thanks heaps.