Users should be able to save their fair predictor to disk and load it back. Without this, they would need to fit a fair predictor every time they want to use one in a new python session, which isn't ideal since it also requires data to fit.
Note that in order to be portable, this needs to be implemented in a smart way, similar to how TabularPredictor's save/load logic works. This would add code complexity to do it right, but it is important for usability.
Sanity tests that should be done when implementing the logic:
[ ] Test fitting fair predictor, saving to location A, then loading from location A, ensuring it works identically
[ ] Test fitting fair predictor, saving to location A, manually moving to location B, then loading from location B, ensuring it works identically
[ ] Test fitting fair predictor, saving to location A, manually moving to location B, then (in a new python session in a new working directory), loading from location B, ensuring it works identically
Users should be able to save their fair predictor to disk and load it back. Without this, they would need to fit a fair predictor every time they want to use one in a new python session, which isn't ideal since it also requires data to fit.
Note that in order to be portable, this needs to be implemented in a smart way, similar to how TabularPredictor's save/load logic works. This would add code complexity to do it right, but it is important for usability.
Sanity tests that should be done when implementing the logic: