betanalpha / jupyter_case_studies Goto Github PK

Do you think stan_utility could be put to pypi?

This would enable users to pip install it. Otherwise this could be installable from git too. Just needs same structure as pypi thing (setup.py + some init)

(I still think all the updates here should go to pystan too).

This would enable you to pin down version and show it where these are used.

Hi,

When I installed pystan the first time I ran into problems with a missing stan_cache.
I solved it by simply changing the code in utils.py

But now I have installed on the institutes server (requested it). Again I meet the same problem, while using compile_model:
No such file or directory: '/home/micmad/.stan_cache/cached-model-c4ed8c57c14f2eabbd4599b428df4a4c.pkl'

I do not have root access, making it hard to manually change the code. Is there a way to specify a location as your cache?

Thanks in advance
Michael

Hi, here are some comments for the stan_utility.py.

python 2

Use from __future__ import division, print_function for py2 compatibility
If division is not imported all the division n/N will result to 0.

precision when printing

In prints use formatter to control the number of digits (e.g. 2)

"{:.2f}%".format(n/N)

_by_chain -function (also see the next subject)

equals to

numpy.swapaxes(ef, 0, 1)

reshaping extract

lines 48 and 49

ef = _by_chain(ef)
ef = ef.reshape(-1, len(ef[0][0]))

equals to

ef = ef.reshape(-1, ef.shape[-1], order='F')

line 53

no need to call extract()

fit.extract().keys() == fit.sim['pars_oi']

line 56

should this line save the result?

shaped[param_name] = shaped[param_name].reshape(*([-1] + dim))

I could find the following functions in the case studies, I am wondering why they are not gathered in any package. So does there exist any Python package with similar diagnostic tools for PyStan?

https://github.com/betanalpha/jupyter_case_studies/blob/master/pystan_workflow/stan_utility.py