search

raptor-ml / raptor

Transform your pythonic research to an artifact that engineers can deploy easily.
https://raptor.ml
Apache License 2.0
148 stars 11 forks source link

[FEEDBACK] How to store Headless features (both in Prod and LabSDK) #208

Open AlmogBaku opened 1 year ago

AlmogBaku commented 1 year ago

Background

How should we store Headless features? and should we?

  1. We MUST have the values for training purposes.
    1. When preparing the FeatureSet: We can run the function against the "key" feature timestamp.
    2. Alternatively, you can run your Headless feature against a DataSource for the sake of "replaying"
  2. What should we do on prod?
    1. should we record the calculations of the "get"?
    2. Alternatively, we might build a special mechanism on the "client-side" that retrieve the FeatureSet(similar to §1.2). This will close the gate over the "simple solution" we have right now for FeatureSet retrieval(= via a view written in SQL)

What do you propose to do?

TBD.

What have you already tried?

No response

What else should we know?

No response