Currently, the num_run is being passed and returned from different functions manually, for example the traditional classifiers are taking a num run and then returning a final num run after it is finished. We should have the backend handle this, which would also allow getting the num run in methods like fit that can be called multiple times. Having the user pass it is not ideal and it is confusing. Also, currently in the fit method we have hardcoded 0 which would lead to overwriting of previous pipelines saved in the disk with the same num run