Compatibility Issue with Python 3.12.6

[Ashour010]

Compatibility Issue with Python 3.12.6

I encountered a JuliaError when trying to fit the PySRRegressor model using the PythonCall package. The error seems to be related to the compatibility of the PythonCall package with Python version 3.12.6.

Steps to Reproduce:
1- Install Python 3.12.6.
2- Install the pysr package using pip install pysr.
3- Run the following code:

import numpy as np
from pysr import PySRRegressor

Sample data

X = np.random.rand(100, 1) # 100 samples, 1 feature
y = 3.0 * X[:, 0] + 2.0 # Example target

Initialize PySR model

model = PySRRegressor(
niterations=100, # Number of iterations
binary_operators=["+", "-", "*", "/"], # Allowed binary operators
unary_operators=["sin", "exp", "log"], # Allowed unary operators
loss="L2", # Loss function (mean squared error)
populations=10, # Number of populations
procs=4, # Number of CPU cores to use
parsimony=0.01, # Encourages simpler equations
maxsize=20, # Maximum size of the equation
)

Fit the model

model.fit(X, y)

View the best equation

print("Best equation:", model.sympy())

View the Hall of Fame (top equations)

print(model)
.
.
Expected Behavior: The PySRRegressor model should fit the data without any errors, and the best equation should be printed.

Actual Behavior: The following error occurs:
JuliaError Traceback (most recent call last)
Cell In[34], line 22
10 model = PySRRegressor(
11 niterations=100, # Number of iterations
12 binary_operators=["+", "-", "*", "/"], # Allowed binary operators
(...)
18 maxsize=20, # Maximum size of the equation
19 )
21 # Fit the model
---> 22 model.fit(X, y)
24 # View the best equation
25 print("Best equation:", model.sympy())

File c:\Users\Administrator\AppData\Local\Programs\Python\Python312\Lib\site-packages\pysr\sr.py:2279, in PySRRegressor.fit(self, X, y, Xresampled, weights, variable_names, complexity_of_variables, X_units, y_units, category)
2276 self._checkpoint()
2278 # Perform the search:
-> 2279 self._run(X, y, runtime_params, weights=weights, seed=seed, category=category)
2281 # Then, after fit, we save again, so the pickle file contains
2282 # the equations:
2283 if not self.temp_equation_file:

File c:\Users\Administrator\AppData\Local\Programs\Python\Python312\Lib\site-packages\pysr\sr.py:1889, in PySRRegressor.run(self, X, y, runtime_params, weights, category, seed)
1886 if isinstance(complexity_of_variables, list):
1887 complexity_of_variables = jl_array(complexity_of_variables)
...
@ PythonCall.JlWrap C:\Users\Administrator.julia\packages\PythonCall\Nr75f\src\JlWrap\module.jl:13
[5] pyjl_callmethod(f::Any, self::Ptr{PythonCall.C.PyObject}, args::Ptr{PythonCall.C.PyObject}, nargs::Int64)
@ PythonCall.JlWrap C:\Users\Administrator.julia\packages\PythonCall\Nr75f\src\JlWrap\base.jl:67
[6] _pyjl_callmethod(o::Ptr{PythonCall.C.PyObject}, args::Ptr{PythonCall.C.PyObject})
@ PythonCall.JlWrap.Cjl C:\Users\Administrator.julia\packages\PythonCall\Nr75f\src\JlWrap\C.jl:63
.
.
.
.
Environment:
-Python version: 3.12.6
-julia version: 1.11.2
-pysr version: 1.3.1
-juliacall version: 0.9.23
-Operating System: [Windows 11 64-bit]
.
.
.
julia> Base.versioninfo()
ERROR: UndefVarError: versioninfo not defined in Base
Suggestion: check for spelling errors or missing imports.
Stacktrace:
[1] getproperty(x::Module, f::Symbol)
@ Base .\Base.jl:42
[2] top-level scope
@ REPL[1]:1
.
.
.
julia> Pkg.status()
ERROR: UndefVarError: Pkg not defined in Main
Suggestion: check for spelling errors or missing imports.
Stacktrace:
[1] top-level scope
@ REPL[2]:1
.
.
.
julia> versioninfo()
Julia Version 1.11.2
Commit 5e9a32e7af (2024-12-01 20:02 UTC)
Build Info:
Official https://julialang.org/ release
Platform Info:
OS: Windows (x86_64-w64-mingw32)
CPU: 16 × AMD Ryzen 7 4800H with Radeon Graphics
WORD_SIZE: 64
LLVM: libLLVM-16.0.6 (ORCJIT, znver2)
Threads: 1 default, 0 interactive, 1 GC (on 16 virtual cores)
Environment:
JULIA_PATH = "D:\Backup\Downloads\julia-1.11.2-win64.exe"
.
.
.
Additional Information: I have verified that the JULIA_PATH environment variable is set correctly and that Julia is accessible from the command line. The issue seems to be specific to the compatibility of the PythonCall package with Python 3.12.6.

[MilesCranmer]

Could you format the code so it is in code blocks, using the "```" syntax? https://docs.github.com/en/get-started/writing-on-github/getting-started-with-writing-and-formatting-on-github/basic-writing-and-formatting-syntax.

[MilesCranmer]

I think the issue is just

loss="L2",

this isn't one of the accepted losses. L2DistLoss() is the correct L2 loss. See https://ai.damtp.cam.ac.uk/pysr/api/#the-objective

[Ashour010]

Thanks for your time and effort, Professor Cranmer. Also, I have a question: can I find one mathematical formula for 9 different parameters influencing each other by linear or nonlinear correlation like G = [A1, A2, A3, B1, B2, B3, C1, C2, C3] by Symbolic Regression?

[MilesCranmer]

Maybe just fit each one in terms of the others, one-by-one? It will try to find simplest formula anyways. Maybe there’s a general dimensionality reduction approach you could try with template expressions though, since it lets you fit multiple at once. Could be cool.