Implement early stopping at predict time using early_stopping_rounds and tolerance which checks if predictions for any samples have changed within the tolerance threshold, and if predictions for such samples do not change for early_stopping_rounds, then those samples are frozen and not passed to TabPFN predict.
Implements model load and save functionality to save generated ensembles and load the same ensembles in a new instance of EnsembleTabPFN. Currently only the pickle format is supported.
Major refactoring involving abstraction of ensemble building from EnsembleTabPFN (similar to data loading functionalities provided in other ML frameworks).
Added tests for EnsembleTabPFN creation and model load and save functionality.
early_stopping_rounds
andtolerance
which checks if predictions for any samples have changed within the tolerance threshold, and if predictions for such samples do not change for early_stopping_rounds, then those samples are frozen and not passed to TabPFN predict.