Closed amesar closed 2 years ago
Import all objects is complex since we have to link the new registered model's version run ID with the already imported experiments and runs. The main problem is scale. If the number of runs can fit into memory, we can use an in-memory hash table of source and destination runs. If not, then what to do? A key value store?
Implemented as a console script import-all which calls import-models since the export directory structure is the same for both.
Implement
Related to: https://github.com/mlflow/mlflow-export-import/issues/11 - Make export_all work correctly