Improving OPR Calculations

Hey everyone. I had a spontaneous thought and figured I’d share it and start this thread for the sake of accurate data this season. If the FIRST and TBA APIs collect data by the robot for both autonomous initiation line crossing and endgame state, then why are we using plain OPR? In my opinion (which has yet to be tested), the best metric should be a combination of these API collections and component OPRs as follows:

Robot 1 Average Auto Init. Line Score + Robot 1 Average Endgame Score + Robot 1 Component OPR for all other scoring**

(Average being adjusted as necessary to reduce regression)

With Endgame likely being the source of a major percentage of match points this year, our OPR calculations will be significantly skewed by these points. If this scoring is available by the robot from the APIs we should use it to our advantage.

Any thoughts on the usefulness of this metric or changes you would make?

This method severely hinders the most important part of OPR: It measures CONTRIBUTION to scoring. A team with a double climb can feasibly have a component OPR of more than 1 climb. A team that consistently fails to climb and causes their partners to fail as well can have a negative number of climbs per match according to their component OPR. This applies for things in auton as well.

By removing some types of OPR from the equation, you lose some of the context in the metric.

1 Like

Makes sense. You bring up a good point. Probably not the best metric for robots in context of alliance partners.

I believe this is what Caleb Sykes tries to quantify with his scouting databases he puts out using the data FIRST provides, but I may be wrong.

2 Likes

You also have to keep in mind that in games like 2019, you might be able to climb onto HAB 3, but you won’t always get the opportunity to depending on your alliance makeup, so even if a team has a HAB 3 climb that works 100% of the time but they only use it in 50% of their matches because their alliance partners can also climb, taking pure data from that is not as strong.

2 Likes

Couple of thoughts:
I’ll be sticking to conventional OPR for my work, primarily for consistency with TBA and other calculators. However, I will calculate rates as well as calculated contributions for both climbing and auto mobility as I’ve done in recent years, so if anyone wants to use those to make variants of conventional OPR such as this one feel free.

I think that every year there are some year-specific adjustments that you can make to OPR to improve its forward looking predictive power (weighting penalty points less than other points comes to mind). Obviously, the best choices for these kind of changes will be the ones we can test and prove to be predictive with real data. We can make educated guesses though.

My prediction would be that swapping auto mobility points with auto mobility rate in an adjusted OPR would improve predictive power. There is basically no interaction between teammates for these points, so the rates just end up being a better way to measure how to get those points. This hypothetical adjustment will probably have close to negligible impact though, as the auto mobility points this year are probably the easiest points we’ve seen since the first rotor in 2017, so most teams will get them in most matches.

Climbing I’m not as confident about, but would still lean toward rates being more predictive than calculated contributions. There’s certainly vastly more interaction between teammates than there is for the auto movement. I think generally though that the increased granularity (one data point for each robot) of rates makes these metrics less noisy and more predictive than calculated contributions.

Don’t personally like the name True OPR. I’m of the opinion that we will never hit a “true” measure of robot skill with metrics, but that doesn’t have stop us from striving for it. :slight_smile: Call it what you want though, when you build metrics you get the naming rights.

3 Likes

Thanks Caleb. Don’t worry, the name will not be sticking around :wink:

I think the match play elements this year that will effect OPRs predictive abilities:

  1. Penalties. A large one time penalty in a match can skew a lot of team’s OPRs. For example: last year at Central New York Qual match 12 had 87 penalty points called. Each member of that alliance saw their OPR increase by about 6 points at the end of quals. Teams that had a qual match with any of those three robots saw their OPR decrease by about 0.7 points. G9/G10/G11/G22/H9/H10 all could add up fast this year.
  2. Power Cell scoring. In 2017 the fuel component OPR for a lot of teams was simply a reflection of the variation in fuel scoring on a per match basis for the good fuel shooting teams. If a top fuel bot missed their auto one match their partners were getting negative fuel OPRs. If the top power cell teams routinely have up and down matches that will hinder OPR’s predictiveness.
  3. Climbing. In 2017 the touch pad components OPR ranged from -25 to 75. There were things you could do to help or hinder partner climb but more often than not it was more a reflection on whether your partners did or didn’t climb in given match. If climbs turn into a more variable thing this year due to strategy or difficulty they could play into OPR.

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.