Hello, is there some tutorial or example on how t...
# troubleshoot
b
Hello, is there some tutorial or example on how to use SDK in order to ingest the MlModel to the datahub?
d
Here is our town hall presentation about MLModel ->

https://www.youtube.com/watch?v=Qo37pcaVkdU

and I think currently you can check the Feast source to check how to use it -> https://github.com/datahub-project/datahub/blob/master/metadata-ingestion/src/datahub/ingestion/source/feast.py
b
Thanks @dazzling-judge-80093 , do I require to have some ingestion source prior for uploading MlModel or can I just upload the MlModel that is not associated with any source like S3, feast, MongoDB etc? What I want to achieve is to upload just the bare model, without any dataset, is it possible?
d
I think if you follow in the source how we do, you should be able to ingest your own one as well. It doesn’t have to be a source.
b
Great, thanks for the info! As always I knew I can count on you. Have a great day!
d
haha, thanks, keep me posted how it goes.
b
@dazzling-judge-80093 so far so good, as far as I understood the core implementation to ingest the MlModel lies within the _get_feature_workunit() function. Although few things are kind of blurry for me, as far as I understood should I create a feature table first in order to create MlModel? Is there some way to just fetch the mlModel file from the disc and let the sdk do it's job by providing the BrowsePaths in aspect?
g
Sure you can do that- there is no requirement for the order of MlModel vs feature table.
Either order is acceptable.
b
@green-football-43791 Is there a way to get mlModel details like hyper parameters and training metrics for MLModelPropertiesClass without using sagemaker for model description?
g
What other tool would you like to use aside from sagemaker? Currently the two ML-related integrations datahub has are Sagemaker and Feast
b
@green-football-43791 well probably the default way without setting those two that you mentioned could be by creating ml_model_urn first and then to fill the properties of this urn by sending the MetadataChangeEvent that contains MLModelPropertiesClass with manually set values by the user or some python script. Although as far as I understood the designed way of uploading MlModel is to use the feast or sagemaker, right?