Could not automatically save pickle file

[leelew]

Hi,

After fitting data by PySRRegressor, I only have two saved files, .csv and .bkup. However, I want to reload all equations after fitting, but I cannot apply from_file because .pkl file is absent. Is there any suggestion?
Thanks!

Regards,
Lu Li

Code is attached:

model = PySRRegressor(
#procs=10,
populations=348,
ncycles_per_iteration=5000,
niterations=1000000000, # Run forever
timeout_in_seconds=int(3600n_hour),
maxsize=50,
complexity_of_operators = {
**{key: 3verylow_complexity for key in medium_complex_ops},
**{key: 2verylow_complexity for key in low_complex_ops},
**{key: 1*verylow_complexity for key in very_low_complex_ops}},
complexity_of_variables=1,
binary_operators=binary_operators,
unary_operators=unary_operators,
extra_sympy_mappings={'inv': lambda x: 1/x},
model_selection="best",
constraints={"pow": (9, 1)},
nested_constraints={"tanh": {"tanh":0}},
loss="loss(x, y) = (abs(max(min(x,100),0) - y))^%d"%loss_exp, #(julia syntax)
temp_equation_file=True,
delete_tempfiles=False,
#cluster_manager="slurm",
#parsimony=10,
bumper=True,
weight_optimize=0.001,)

[leelew]

Hi,

I found from_file could directly load equations from .csv file. Thus my problem is solved.

I have a further question. I want to count free parameters of equations instead of directly use complexity. Is there any method to automatically get the number of free parameters? I have to manually count it for all equations after fitting data.

Thanks!

Regards,
Lu Li

[MilesCranmer]

Want to make a new thread for the constants?

And great to hear you could fix it with the .csv file loading!

[leelew]

I try to count constant in the resulting equations using sympy library. However, I did not know how to count it. e.g., $1/(ax+b)$. Is there any suggestions?

[leelew]

I found use sympy could count constants directly by following codes:

import sympy as sp
for i in range(len(model.equations_)):
    expr = model.equations_.sympy_format.iloc[i]
    print(expr)
    constants = [term for term in expr.atoms(sp.Number)]
    constants = np.setdiff1d(constants, [-1])
    print(len(constants))