OPR vs Strength of Field

I imagine that this has already been discussed somewhere. Can someone refer me to the proper thread where OPR is considered in light of “Strength of Field” (How strong the opponents are at the event).

Watching our OPR, I can see it decrease as the strength of opponents increases in the fields. I was wondering what I might do in my calculations to account for that? Or if it is only accountable in hindsight.

How are you planning to use OPR? It’s likely not necessary.

We go to 2 or 3 regionals a year plus (usually) Houston. Usually a Week1, Week 4 or 5… So, for the first few matches of the day, I need to prepopulate some kind of estimate on their scoring capability to estimate how they will do in the match. We’ve been using ELO factored against an arbitrary scoring cap. I think it would be smarter to use some kind of weighted average of OPR’s.

Anyway, that’s the general idea.

In 2019, our estimated alliance scores were +/- 10%. In 2020, that improved to +/- 7%. I’d like to get it down to +/- 5%. (Those are rough estimates) Call me crazy…

1 Like

I don’t want to poopoo the idea too quickly, but I’d caution giving OPR more predictive power than it really has. Understandably, you’ve got to work with the data you can get to get the intel you want. If you’re just trying to figure out “yeah, we’ve got a pretty tough match coming up”, then sure, maybe it’s predictive enough. You’re probably better off doing some pre-event scouting.

5 Likes

Here is the most similar thread I can think of off the top of my head: looking at OPR across events

It’s more general than very technical though. My personal take on strength of field effects on OPR is that it is extremely game dependent. In some years, robot A competing at a weaker event will have a higher OPR than the same robot A competing at a stronger event, while in other years the opposite will be true.

The best thing you can probably do would be to pull out the components of OPR that are largely event-strength independent and then add an event adjustment to the other component OPRs based on the strength of their previous event. Now the exact components to do this for will vary based on the game, using 2020 as an example I would say you could comfortably pull out auto scores as being event independent, and then put an event-strength adjustment on teleop/endgame scores (do whatever you want with penalties, they are basically noise anyway). You could calculate such an event adjustment by comparing pre-event predicted component OPRs to actual post-event component OPRs to see if stronger events tend to have higher actual component OPRs than predicted.

This is all hypothetical, I’ve never done this kind of correction before but I think it at least has some sound theory behind it.

I’d encourage you to enter into my prediction contests whenever events start up again. Your predictions as they currently stand would be a very strong contender. They sound a bit too good to be true to me (or they come from very small sample sizes) but I’d love to be proven wrong.

1 Like

That would depend on 1) how well I am remembering the statistics I was getting and 2) our ability to maintain good scouting standards. I’d also say 2019 was a really easy year to do predictions (and scouting). Many years have been much harder. I’ve only been doing scouting and predictions for 3 years. So I don’t have the historical data to prove anything to anyone’s satisfaction.

We do things like

given the predicted outcome, who does our alliance defend, what type of defense do we put on them, and who do we assign the defense to.

Who will the opposing alliance be choosing to defend and can our alliance counter defend effectively.

Like that.

I really want to know how close going to be and how desperate each alliance is going to be.