Hi @ItsIronOxide,

thanks for opening this issue. It really helps though, if you can provide a minimal reproducible including input data. That doesn't need to contain anything proprietary, but rather be a scheme of your way to use scikit-learn models. The only thing that counts is that we see what you did and can re-create the problem.Thank you.

In addition to what Stefanie said, you mention unpickling estimators from older scikit-learn versions. https://scikit-learn.org/dev/model_persistence.html#replicating-the-training-environment-in-production could be helpful for you - basically: it might work, or maybe not. You should get a specific InconsistentVersion warning to tell you.

Without a code snippet showing what you are trying to do and the resulting traceback it is unfortunately nearly impossible to give concrete advice.