Sample notebooks

[paravatha]

Hey all, Glad to see this initiative. Earlier this year, I tried Feast 0.9 on GCP/GKE : https://paravatha.medium.com/feast-setup-your-own-ml-feature-store-on-kubernetes-5b3193c2b62c

Unfortunately, I don't have access to create Redis. So, If you have any sample notebooks for just the offline store part, I'd like to give it a try.

[samuel100]

@paravatha thanks for the feedback! So you have access to a MS SQL Store (e.g. SQL Server, Azure SQL DB, Synapse) but you cannot create an Azure Cache for Redis?

We can get some more samples added in.

[paravatha]

@paravatha thanks for the feedback! So you have access to a MS SQL Store (e.g. SQL Server, Azure SQL DB, Synapse) but you cannot create an Azure Cache for Redis?

We can get some more samples added in.

Thanks @samuel100. I am planning to try this on Azure Sandbox playground on ACG, "Redis Cache" is not available there https://help.acloud.guru/hc/en-us/articles/360001389296

[samuel100]

@paravatha thank you for the context!

If you remove the online_store from the feature_store.yaml file, so that it looks like:

registry:
  registry_store_type: feast_azure_provider.registry_store.AzBlobRegistryStore
  path: https://<ACCOUNT_NAME>.blob.core.windows.net/<CONTAINER_NAME>/<DIRECTORY>/registry.db
project: production
provider: feast_azure_provider.azure_provider.AzureProvider
offline_store:
    type: feast_azure_provider.mssqlserver.MsSqlServerOfflineStore
    connection_string: ${SQL_CONN}

It will create a local SQL lite instance for the online store, which will work for the training and materialization parts of the tutorial. Real-time inference is tricky without an online store. We could potentially use SQL DB as an online store for these sandbox scenarios -- I will investigate.

[tarockey]

Regarding the sample notebooks - when trying to test the deployment to ACI, the example model_service_env.yml references feast-azure-provider==0.1.0 as a pip package, and the Image fails to build. I assume this example is pending a push for this package to be included in pip, correct?

[samuel100]

@tarockey We have just release the package to PyPi.

[samuel100]

This is now a duplicate of #31 since Cosmos DB is available in the Azure Sandbox.