- Set of Databricks notebooks to perform all MLflow export and import operations.
- You use these notebooks when you want to migrate MLflow objects from one Databricks workspace (tracking server) to another.
- The notebooks in the git directory are generated by using the Databricks GitHub version control feature.
- You will need to set up a common cloud location mounted on DBFS in your source and destination workspaces.
- See the notebook README.
Columns
- Notebook - name of notebook.
- git import - Databricks github sync format.
- HTML - Viewable Convenience format. Note that the widgets are not displayed.
Notebooks
| Notebook | git import | HTML |
|---|---|---|
| Export_Run | link | link |
| Import_Run | link | link |
| Export_Experiment | link | link |
| Import_Experiment | link | link |
| Export_Model | link | link |
| Import_Model | link | link |
| Common | link | link |
| _README | link | link |
In order to import the notebooks into your Databricks workspace, use the workspace import_dir and workspace import Databricks CLI commands.
databricks workspace import_dir git /Users/me@mycompany.com/mlflow-export-import
databricks workspace import --language PYTHON git/_README /Users/me@mycompany.com/mlflow-export-import/_README
The separate _README import is needed since there is apparently a glitch in that when the _README file is checked into git, a .py extension is not added.
Push the wheel library to DFBS.
python setup.py bdist_wheel
databricks fs cp \
dist/mlflow_export_import-1.0.0-py3-none-any.whl \
dbfs:/home/me@mycompany.com/lib/wheels/mlflow_export_import-1.0.0-py3-none-any.whl
Then attach the library to your cluster.