Lockin scikit-learn version and estimators version

[luigif2000]

Dear, thank again for Your wonderfull job; I have this problem:

1) time ago, I fit thousands models with flaml (environment scikit-learn version 1.0.2, xgboost some few version older) 2) saving model with pickle (as suggested in flaml documentation) 3) upgrade scikit-learn 1.0.2 -> 1.2.0 (for conda suggest/performance) 4) upgrade xgboost to latest version (for conda suggest/performance) 5) upgrade flaml to 1.1.1 (for the new models , replacing the olds one)

now when I load back with pickle the old models i have blocking errors: ERROR 1) : scikit-learn version mismatch -process stop ERROR 2) : xgboost version mismatch--process stop

is there workarounds to solve this kind of versions lockin?

Thanks in advance and waiting Your kindly reply.

luigi

[thinkall]

Hi @luigif2000 , have you tried to downgrade scikit-learn and xgboost after installing the latest flaml?

[luigif2000]

Hi, sorry for the late due to Covid problems ; thanks so much for your kindly reply. I was thinkinging to better have the last scikit-learn and xgboost sinceresly. I'm replacing the old xgboost models with the new one...and hope it will better.

thanks again and best regards.

luigi