Replies: 2 comments
-
Hi @wyli , Thanks for raising the question. Thanks. |
Beta Was this translation helpful? Give feedback.
-
Where an imported package is used isn't easy to test, we could create a virtual environment with a github action and install everything specified then try to instantiate every object in one of the scripts to see if it's possible, this would catch the lack of pytorch-ignite for example when instantiating any engine type. This would miss cases where the package is used in code that isn't run until train/inference time. A more in-depth method would be to run static code analysis to collect together all instances of package importation to get a list of all dependencies, this would omit only the most bizarre cases of |
Beta Was this translation helpful? Give feedback.
-
Related to this ticket Project-MONAI/MONAI#4456,
the metadata json should also include the info about pytorch-ignite version. For example this bundle clearly depends on ignite, but it's not described in the metadata:
model-zoo/models/spleen_ct_segmentation/configs/metadata.json
Lines 11 to 13 in 0ded5a4
I guess in general the question is how we make sure the stated
optional_packages_version
is indeed the minimally required set of packages to run a bundle.Beta Was this translation helpful? Give feedback.
All reactions