[BUG]: Compatibility issues with model loading and TensorBoard

[jackeuylov]

What happened?

A bug happened!
When I am using model loading and TensorBoard simultaneously, I get the following error:

The case used is the code in Examples：
first, Using a model with TensorBoard to generate hall_of_fame：

import numpy as np
from pysr import PySRRegressor, TensorBoardLoggerSpec
rstate = np.random.RandomState(42)
X = rstate.uniform(-3, 3, (1000, 2))
y = np.exp(X[:, 0]) + X[:, 1]
logger_spec = TensorBoardLoggerSpec(
    log_dir="logs/run",
    log_interval=10,  # Log every 10 iterations
)
model = PySRRegressor(
    binary_operators=["+", "*", "-", "/"],
    logger_spec=logger_spec,
)
model.fit(X, y)

then, Using the model loading function to optimize the hall_of_fame.

import numpy as np
from pysr import PySRRegressor, TensorBoardLoggerSpec
rstate = np.random.RandomState(42)
X = rstate.uniform(-3, 3, (1000, 2))
y = np.exp(X[:, 0]) + X[:, 1]
model = PySRRegressor.from_file(run_directory="outputs/20241212_222454_uKch8W")
model.fit(X, y)

Then there will be the above error.

Version

1.1.0

Operating System

No response

Package Manager

None

Interface

IPython Terminal

Relevant log output

No response

Extra Info

No response

[MilesCranmer]

Thinking out loud, I think I know the issue. It’s the fact we don’t remove the logger from the state saved to the pickle file. Basically, we can’t pickle Julia objects. This new feature needs to be masked from the pickle

[MilesCranmer]

#780 should fix this