Should Shepherd-ARAX be inspecting the MetaKG?

[saramsey]

Max Wang raised this point in today's Engineering Committee meeting. He writes:

this might be a more involved conversation, but with the move to Retriever and it being an abstraction on top of all KPs, I wonder if it would make sense for ARAX to remove it’s metakg checking entirely? Retriever will already be choosing which KPs it should send the query to (in Tier 1)

What do we think?

[edeutsch]

Feels icky. I think they should publish a decent MetaKG. But convincing them to publish a decent MetaKG would likely be a long drama, so I'm fine with just putting in a hack in ARAX to always proceed to Retreiver no matter what the MetaKG says.

[saramsey]

I've created a branch issue-2598 off of the issue2521 branch, for us to work on the proposed hack.

[saramsey]

See 4bd99ae, my attempt at a patch (in kp_selector.py)

[saramsey]

I have rolled out my patch to arax.ncats.io/shepherd. I think it is working? I see one edge coming from retriever, on the Example 2 query ("What drugs treat Castleman disease?"):