Contains the SAME project's python codebase.
The SDK relies on the conda package to support Conda environments. Conda's setup.py is incompatible with python ^3.10, however, so we have vendored the package directly in the vendor directory.
The pipreqs tool is used to automatically suggest missing dependencies to the user if they have unresolved imports in their notebook. pipreqs doesn't have an importable library, however, so we have also directly vendored the code.
Do not make changes to vendored code, unless you are pulling in upstream commits from the conda or pipreqs repositories.
The most extensively developed backend right now is kubeflow, which supports the following set of features:
- Notebooks containing multiple steps.
- Configuration with a specified docker image to run against.
- Configuration with a
requirements.txtof dependencies to install. - Integration of
datasetsconfiguration and thesame.datasets(...)SDK.
Other backends support the following subset of these features:
| 1 | 2 | 3 | 4 | |
|---|---|---|---|---|
| kubeflow | x | x | x | x |
| durable functions | ||||
| aml | ||||
| vertex | ||||
| pachyderm | x | x |