NotImplementedError with pymc_extras.fit_laplace and other

[VictorBaranovMain]

Good evening!

I've been interested in this package long ago, and tried to use it in previos versions, however long before it did not suited well for purporses of my reserach. Now it is great, but I've faced problem (or multiple of them, actually)

When I try to "run all" examples - in estimation section I get multiple "NotImplementedError"'s when use fit_laplace. Some same staff when using NUTS sampler - but if fixes itself with setting " nuts_sampler="pymc" ".

Plus, when I am try to run file "fit_rbc_to_us_data.ipynb" I get error of

"
RuntimeError: Coordinate value has incorrect type

Caused by:
RuntimeError: Could not convert to Value. Unsupported type.
"

when I evaluate cell with " import nutpie as ntp ". May be there is more, but I've not checked others ipynb from case_stuty folder.

All of theese error occur both when I use own created .env with installing gEconpy, with .env, then installing pytensors and then installing gEconpy, with .conda created with .yml file - every time same error with fit_laplace. Which version of pytensors is supported? Or is it depends on pymc_extras version? Or may be I am doing smth wrong :D
Thank you!

[jessegrabowski]

Hey hey!

Thanks for giving the program a try. There has been a lot of upstream development work in pymc, pytensor, and nutpie recently. Let me go re-run all the examples and make sure things are in working order. I'll help you debug if it's on your end, but I have a hunch there might be some subtle bugs to quash.

[VictorBaranovMain]

Thanks! I'll wait for any updates.

p.s. I will try something more with repo and enviroments, hope I'll succed)

[jessegrabowski]

The first thing to try on the environment is to downgrade nutpie

[jessegrabowski]

Following up on this -- I reran all the examples locally without any issues. Can you post the output of conda list so I can help you debug your environment?

Sorry I spoke too soon. I can reproduce the laplace error, that's a bug. The MCMC methods all work though. What other errors were you hitting? I'd still like if you could share your conda environment

[VictorBaranovMain]

Thanks for help!
Well, I can not understand what happened actually. Becase when I wrote an issue - I han not only laplace problem, but also could not run NUTS in any form. And back them I used your conda_env/enviroment.yml file. And problems where in many places.

This morning I just deleted the enviroment and created it again. And, aside from laplace error & problem with forkserver parameter in pm.sample (it is not supported on Windows) - I faced none of that problems, even with nutpie.

So, sorry on that part - probably smth happened in that time or there were some issues with conda. But it was created the same way as tommorow's one .__.

as a result I suppose only Laplace bug is left.

p.s. by the way, do you know any easy way to convert from .mod model into .gcn except manual writing?)

[jessegrabowski]

Yeah python packaging is horrible black magic. Glad the errors cleared out.

For converting .mod -> .gcn I don't have a tool yet, but if you open an issue to ask for one it would be on the roadmap :)

[VictorBaranovMain]

Anyway, thanks for helping with issue. I gess I'll leave it open untill Laplace bug will be fixed.

Overall - thanks for package developing! Great work)
Hope I'll be aible to contribute to developing in the future as my skills will suit for it.