Working with pre-compilation

[AKAL4nka]

I'm having issues with pre-compilation not working causing me to re-compile every time I run, which is super slow.
I'm getting a warning:
.....Compiling Julia backend... /home/usr/.local/lib/python3.10/site-packages/pysr/julia_helpers.py:217: UserWarning: Your system's Python library is static (e.g., conda), so precompilation will be turned off. For a dynamic library, try using pyenvand installing with--enable-shared: https://github.com/pyenv/pyenv/blob/master/plugins/python-build/README.md#building-with---enable-shared.

Could this be improved?

[MilesCranmer]

Hi @AKAL4nka,

The issue is in the use of a Python binary which has a static libpython (such as the one installed by conda), so PyJulia can't make use of Julia's precompilation artifacts. However, do note that if you simply keep the same Python instance open, the second time you run model.fit, it won't need to do the compilation again. The compilation is only the first call to model.fit in Python.

I recommend using autoreload https://ipython.org/ipython-doc/3/config/extensions/autoreload.html if you are working on a python library – it will automatically load changes to python code – so you don't need to keep restarting it.

Alternatively, if you are open to helping, there is information here about eliminating this issue altogether: JuliaPy/pyjulia#496 – which would let you actually use the precompilation on static libpython. I think it won't be too bad to get it through (depending on your software dev background); it just needs someone to actually do it!

Alternatively, the other option is to switch to juliacall rather than pyjulia for calling Julia code. This won't be too hard to do, the main issue is just that there's no conda-forge feedstock of juliacall yet so people wouldn't be able to get PySR on conda.

Cheers,
Miles

[MilesCranmer]

@AKAL4nka on the latest version of PySR this will no longer be an issue.

[AKAL4nka]

Awesome!