Can deployment be even more granular

Question

It would be amazing if targeted deployment could be even more granular or requested on-demand. For instance, being able to download an ML model tailored to a certain location being visited would be very powerful.

Any chance of that being in the pipe-line or better yet, is there a way to do that already?

678

Posted by

CandideApixel

Reply

Add a Comment

Answer 1

As @kerfuffle mentioned, that link does require some additional management, but it is still a perfectly valid way to retrieve models remotely.

For Core ML Model Deployment, we do not natively support targeting based on location. We provide targeting based on the language and region code of the device, but this is user configured and may not reflect their actual location.

Note that we do support requesting "on-demand". The initial request for a model collection using MLModelCollection.beginAccessing will go and fetch the specified collection, with the handler called upon completion. One option for your use case would be to define multiple collections for certain locations. Then you can request the device's location in your app at runtime (after being granted permission), and begin accessing the collection best suited for the result.

Posted by

Frameworks Engineer

Add a Comment

Answer 2

Ha. I've answered my own question:
https://developer.apple.com/documentation/coreml/core_ml_api/downloading_and_compiling_a_model_on_the_user_s_device

Posted by

CandideApixel

Add a Comment

Answer 3

The difference here is that you need to host the model yourself on a server somewhere. It's not automagically handled for you like the new deployment stuff is.

Posted by

kerfuffle

Add a Comment

Answer 4

As @kerfuffle mentioned, that link does require some additional management, but it is still a perfectly valid way to retrieve models remotely.

For Core ML Model Deployment, we do not natively support targeting based on location. We provide targeting based on the language and region code of the device, but this is user configured and may not reflect their actual location.

Note that we do support requesting "on-demand". The initial request for a model collection using MLModelCollection.beginAccessing will go and fetch the specified collection, with the handler called upon completion. One option for your use case would be to define multiple collections for certain locations. Then you can request the device's location in your app at runtime (after being granted permission), and begin accessing the collection best suited for the result.

Posted by

Frameworks Engineer

Add a Comment

Answer 5

Excellent. Collections do look promising.

Thank you.

Posted by

CandideApixel

Add a Comment

Can deployment be even more granular

Accepted Reply

Replies