Fix #62

[aseyboldt]

Seems the coords can be None in pymc...
fixes #62

[twiecki]

don't we have ADVI at the pytensor level in pymc already?

[aseyboldt]

Sorry, this wasn't supposed to end up in the PR. I rebased

[twiecki]

_______________________________ test_pymc_model ________________________________

    def test_pymc_model():
        with pm.Model() as model:
            model.add_coord("foo", length=5)
            pm.Normal("a", dims="foo")
    
>       compiled = nutpie.compile_pymc_model(model)

tests/test_pymc.py:23: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/nutpie/compile_pymc.py:191: in compile_pymc_model
    coords = {name: pd.Index(vals) for name, vals in model.coords.items()}
/opt/hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/nutpie/compile_pymc.py:191: in <dictcomp>
    coords = {name: pd.Index(vals) for name, vals in model.coords.items()}
/opt/hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/pandas/core/indexes/base.py:524: in __new__
    raise cls._raise_scalar_data_error(data)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

cls = <class 'pandas.core.indexes.base.Index'>, data = None

    @final
    @classmethod
    def _raise_scalar_data_error(cls, data):
        # We return the TypeError so that we can raise it from the constructor
        #  in order to keep mypy happy
>       raise TypeError(
            f"{cls.__name__}(...) must be called with a collection of some "
            f"kind, {repr(data) if not isinstance(data, np.generic) else str(data)} "
            "was passed"
        )
E       TypeError: Index(...) must be called with a collection of some kind, None was passed

/opt/hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/pandas/core/indexes/base.py:5265: TypeError

[twiecki]

Processing ./dist/nutpie-0.9.1-cp310-cp310-manylinux_2_28_x86_64.whl
INFO: pip is looking at multiple versions of nutpie[all] to determine which version is compatible with other requirements. This could take a while.
ERROR: Could not find a version that satisfies the requirement pyarrow>=12.0.0 (from nutpie[all]) (from versions: none)
ERROR: No matching distribution found for pyarrow>=12.0.0

[maresb]

I've never used --find-links before, but this feels like it may be overly-complicated.

I would have done something like

pip install --force-reinstall dist/nutpie*.whl

but maybe this doesn't work cross-platform?

[maresb]

Not sure why this would fail on OSX but succeed on Linux:

=================================== FAILURES ===================================
_______________________________ test_pymc_model ________________________________

    def test_pymc_model():
        with pm.Model() as model:
            model.add_coord("foo", length=5)
            pm.Normal("a", dims="foo")
    
>       compiled = nutpie.compile_pymc_model(model)

tests/test_pymc.py:23: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../../hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/nutpie/compile_pymc.py:191: in compile_pymc_model
    coords = {name: pd.Index(vals) for name, vals in model.coords.items()}
../../../hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/nutpie/compile_pymc.py:191: in <dictcomp>
    coords = {name: pd.Index(vals) for name, vals in model.coords.items()}
../../../hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/pandas/core/indexes/base.py:524: in __new__
    raise cls._raise_scalar_data_error(data)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

cls = <class 'pandas.core.indexes.base.Index'>, data = None

    @final
    @classmethod
    def _raise_scalar_data_error(cls, data):
        # We return the TypeError so that we can raise it from the constructor
        #  in order to keep mypy happy
>       raise TypeError(
            f"{cls.__name__}(...) must be called with a collection of some "
            f"kind, {repr(data) if not isinstance(data, np.generic) else str(data)} "
            "was passed"
        )
E       TypeError: Index(...) must be called with a collection of some kind, None was passed

../../../hostedtoolcache/Python/3.10.13/x64/lib/python3.10/site-packages/pandas/core/indexes/base.py:5265: TypeError

[aseyboldt]

Thanks for having a look!

The test failure is I think because it seems to be using the nutpie from pypi instead of the new one. I don't quite understand why that would happen on mac but not on linux though...

[aseyboldt]

Ah, I think we need the separate install step in the other ci jobs as well...

[aseyboldt]

 pip install --force-reinstall dist/nutpie*.whl

I think the problem with this is that there are several whl files, for different python versions.