Returned feature vector when on demand feature views are specified output dependent features

[adchia]

When specifying on demand feature views at retrieval time (e.g. get_X_features), the output feature vectors include e.g. request data or dependent feature vectors, even if users did not specify said features.

Expected Behavior

Non-specified dependent feature values are not returned in output

Current Behavior

Non-specified dependent feature values are in output

Steps to reproduce

Specifications

• Version:
• Platform:
• Subsystem:

Possible Solution

[felixwang9817]

Resolved by #2004

[adchia]

This isn't resolved yet for get_historical_features

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[tmihalac]

Hello, I am new to Feast can you please explain to in detail what is needed to fix this for get_historical_features ?

This is what Tornike Gurgenidze answered in Slack : "Hi, most of the code you might be interested in is in RetrievalJob class (https://github.com/feast-dev/feast/blob/master/sdk/python/feast/infra/offline_stores/offline_store.py#L61), look for to_df and to_arrow methods. When individual feature store implementations create subclasses of RetrievalJob, they normally override _to_df_internal and _to_arrow_internal methods only. to_df then calls _to_df_internal , applies final logic to calculate values for odfvs and returns the final output.

Also, I might be totally wrong about this, but in case of get_historical_features I don't think the bug is that extra unspecified feature values are being returned, but rather that if the user doesn't specify all the features that requested odfv features depend on, RetrievalJob will be unable to calculate feature views and the call will most likely fail. offline_store.py class RetrievalJob(ABC):"

I need further explanation about it and also how to reproduce the bug to be able to test the fix.