Improve model saving in the sklearn API to ensure future compatibility and safety. Currently, we save the entire object using torch.save, which may cause issues with different CEBRA versions. This pull request (PR) tackles this problem by suggesting a solution that involves saving only the model weights and all the arguments used, allowing for complete reconstruction of the CEBRA object at loading time.
Improve model saving in the sklearn API to ensure future compatibility and safety. Currently, we save the entire object using torch.save, which may cause issues with different CEBRA versions. This pull request (PR) tackles this problem by suggesting a solution that involves saving only the model weights and all the arguments used, allowing for complete reconstruction of the CEBRA object at loading time.
Fix https://github.com/AdaptiveMotorControlLab/CEBRA-dev/pull/643 Fix https://github.com/AdaptiveMotorControlLab/CEBRA/pull/33 Fix https://github.com/AdaptiveMotorControlLab/CEBRA-dev/issues/623