search

RobokopU24 / Feedback

Feedback on the ROBOKOP project
https://robokop.renci.org
0 stars 0 forks source link

Reconsider approach to subclassing? #185

Open karafecho opened 8 months ago

karafecho commented 8 months ago

This issue is to suggest that we reconsider our approach to subclassing.

As I've mentioned a few times, I have mixed feelings about subclassing. Subclassing not only introduces performance issues, as made evident while troubleshooting #184, but it also just muddies the results with more-or-less redundant information and ontological linkages/inferences that aren't terribly interesting, imo.

I feel like the ideal approach is to allow a user to turn on/off subclassing. For instance, I think a user might be interested in subclassing after first exploring results that precisely matched the q-graph query, i.e., an iterative exploration process.

Evan and I discussed this a bit, and I think we agree that subclassing sounds great and likely has use cases that would benefit from it, but it also adds a lot of complexity and introduces performance issues that together suggest questionable benefits.

cbizon commented 8 months ago

I think that this is an interesting point. It seems to me that there are 2 independent concerns:

  1. Do we want subclassing in the first place
  2. At the times it is appropriate, how can we make it efficient

I think that the answer to 1 is yes, because I think you will miss answers you want when you do longer queries without it. But this may be wrong. So I wonder if we should try to come up with some test queries to try out and base this on some level of actual results.

2 is trickier maybe - I'd like to understand better why subclassing makes things so slow before I give up :)

cbizon commented 3 months ago

@karafecho The original comment here you say "it also just muddies the results with more-or-less redundant information and ontological linkages/inferences that aren't terribly interesting, imo." Can you give some examples? It's sort of hard to know if I agree or not....

karafecho commented 3 months ago

I also have mixed feelings about subclassing. Here's a very simple example that triggered my doubts about subclassing.

image

This is from a query I ran for Scott Auerbach's use case. The Monarch edges add to the "story", for sure, but they don't really provide someone like Scott with any new knowledge.

My suggestion is to make subclassing optional by allowing users to turn it on/off.