You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
a generic "python script entry points" similar to what setup(entry_point=...) or poetry.scripts expect
(e.g.: my_package.my_module:some_function or my_package.my_module:SomeClass.a_method)
a URI based reference to a script or executable
(somewhat relates to Add a Software object #29 about the href requirement)
(for this, I would suggest that "expression": {"href": "<uri>", "type"; "<media-type>"} is used, since expression allows anything)
some docker container or similar, using an expression like ghcr.io/NAMESPACE/IMAGE_NAME:latest (or a more explicit docker run call)
The processing extension initially aims at tracing the processing lineage of the item. I understand you want to specify possible processing on the item? The expression object is left free to add any data structure that would represent the processing applied. If you want to provide with more processing format and related convention, feel free to provide them with examples.
Reacted by Matthias Mohr and Francis Charette-Migneault
I agree. If the above suggestions are valid, I can open a PR to add the examples.
Users are always free to define whichever format that suits their needs, but if we push them toward a common strategy with good examples, it can help "standardizing" some work later on.
Also related, once the Machine Learning Model (MLM) extension will be released, I plan to suggest another example similar to the following for STAC Items representing a derived product resulting from model predictions (eg: land-cover).
Activity
emmanuelmathot commentedon Apr 4, 2024
The processing extension initially aims at tracing the processing lineage of the item. I understand you want to specify possible processing on the item? The expression object is left free to add any data structure that would represent the processing applied. If you want to provide with more processing
formatand related convention, feel free to provide them with examples.fmigneault commentedon Apr 4, 2024
I agree. If the above suggestions are valid, I can open a PR to add the examples.
Users are always free to define whichever
formatthat suits their needs, but if we push them toward a common strategy with good examples, it can help "standardizing" some work later on.Also related, once the Machine Learning Model (MLM) extension will be released, I plan to suggest another example similar to the following for STAC Items representing a derived product resulting from model predictions (eg: land-cover).
{ "processing:level": "L4", "processing:expression": { "format": "stac-mlm", "expression": { "href": "<URI-to-MLM-STAC-Item>", "type": "application/geo+json" } } }As described in https://github.com/crim-ca/dlm-extension/blob/validate/best-practices.md#processing-extension
add more processing:expression examples for python, docker and generi…