mml.core.scripts.schedulers.upgrade_scheduler

class UpgradeScheduler[source]

AbstractBaseScheduler implementation for the Dataset and MML Upgrade process. Includes the following subroutines: - upgrade - downgrade For upgrade we always assume to upgrade to the currently installed version of MML. For downgrade, we assume that previously MML has been upgraded to the currently installed version and is now downgraded to the specified version in cfg.mode.version.

__init__(cfg: DictConfig)[source]

create_routine() → None[source]

This scheduler implements two subroutines, one for dataset preparation and one for task preparation.

Returns:: None

prepare_exp() → None[source]: Prepare experiment expects tasks to be present and loads these into task factory container. Here this should be avoided (including after_preparation_hook()).

upgrade_0_12() → None[source]: This performs the necessary updates to results and data from 0.11 to 0.12 version of mml-core. Iterate over all installed tasks and update the keys: tags, train_tuples, unlabeled_tuples, test_tuples