Add support for stac:mlm extension

[jonas-hurst]

STAC Machine Learning Model (stac:mlm) extension is currently not supported.

A minimal implementation is provided here, but it feels more like a prototype, e.g. @property methods missing in classes.

I am wondering if there is a specific reason that a proper implementation is not included in pystac yet? Maybe @fmigneault or @rbavery have a good reason it is not in here yet?

I am happy to submit a PR with an implementation.

[jsignell]

I am not sure if there is a reason other than pystac extension classes are annoying to write. In case that is the reason I want to mention that @gadomski has started laying out a pystac rewrite (#1524) that will make writing extension classes a whole lot simpler. The new way of defining an extension class will probably involve a bit more than this: https://stac-utils.github.io/pystac/api/extensions/projection/ but simplicity is the goal!

[rbavery]

Glad you're interested in using stac:mlm!

It felt easier to me to use pydantic, so yeah the main reason is that pystac just generally seemed too complicated and I preferred to implement in pydantic.

Implementation in pystac is welcome as long as it conforms to the MLM schema, but I think it'd be easier to use stac-model for MLM metadata validation.

This example shows how.

[fmigneault]

As mentioned by @jsignell, I heard about that rewrite some time ago during a STAC biweekly meeting. I'm holding off further implementation until this refactor is better defined since it is indeed troublesome to write the extension at the time being.

For reference, I did try to implement some generic "helper" class that combines a pydactic model (eg: what stac_model offers) with the pystac extensions classes here: https://github.com/crim-ca/stac-populator/blob/ea774c5b8f1b8f8592276cb5c0c4a944a2b8ceca/STACpopulator/extensions/base.py#L68

These can be used for some STAC extension implementations that are not provided by pystac such as cmip6:

• pydantic propertes: https://github.com/crim-ca/stac-populator/blob/ea774c5b8f1b8f8592276cb5c0c4a944a2b8ceca/STACpopulator/extensions/cmip6.py#L81
• pystac: https://github.com/crim-ca/stac-populator/blob/ea774c5b8f1b8f8592276cb5c0c4a944a2b8ceca/STACpopulator/extensions/cmip6.py#L208-L212
• and their many combinations for each STAC Item/Asset/ItemAssets/Collection/Summaries: https://github.com/crim-ca/stac-populator/blob/ea774c5b8f1b8f8592276cb5c0c4a944a2b8ceca/STACpopulator/extensions/cmip6.py#L278-L407

All this is what inspired the example MLM code linked by @rbavery

But yeah, using them isn't a great user experience, and it's hard to debug when things go wrong...
The refactor will be welcome.

[jonas-hurst]

Thanks for all your input.

I decided to provide an implementation for the current version of pystac as it is beneficial for a current project of mine.