@Aditya Kotiyal ,

ML models are routinely saved and fetched to/from CDF files in Functions to enable end to end use cases..  Whether CDF Functions are appropriate to ‘train’ and ‘retrain’ depends on the use case as CDF Functions have finite resource limits (memory, cpu and runtime).  

-Jason

fyi @Anvar Akhiiartdinov @Andris Piebalgs @Chad Hutchison Can you advise further?

Thanks @Jason Dressel .

Possible to share those resource limits?

@Aditya Kotiyal Cognite Functions resource limits can be found here.

Hi @Aditya Kotiyal,

We are following up to see whether you're satisfied with the responses you've received.

If you found the responses offered by our community members to be instructive and helpful, we kindly request that you consider marking the most helpful response as the "Best Answer."

Best regards,
Dilini 

Hi @Dilini Fernando ,

I was expecting other tagged by Jorgen to reply.

It will be good to understand how much CDF can be leveraged for AI/ML kind of workflows.

Hi@Anvar Akhiiartdinov @Andris Piebalgs @Chad Hutchison

Please let us know your thoughts on this.

Br,

Dilini

@Aditya Kotiyal ,

I worked with SWN team on this.  In short

• Build/Train model externally → Push to CDF as a File
• Within a Cognite Function, load the ML model and call it
• Retraining can be via Cognite Function as well (within CDF Function limitations: memory, runtime) → push back to CDF as a File

Hope this helps,
-Jason

Thanks @Jason Dressel