OPR for Power Up

Gamma · January 8, 2018, 3:23pm

What scouting metrics will be useful this year?

Considering that the majority of points scored this year will be dependent on the ability (or inability) of opponents, I would guess that OPR will not prove itself as useful. I’m not very familiar with DPR, will it work with the scoring scheme?

Brian_Maher · January 8, 2018, 3:24pm

Scouting data >> OPR almost every time. Good scouting data is tells the full story, and OPR isn’t even meaningful for many games.

aronsa · January 8, 2018, 3:33pm

Yes, because of the time-based points OPR will be especially unhelpful. Scores will generally be tighter.

Daniel_LaFleur · January 8, 2018, 3:37pm

Cubes per match, where the cubes are scored, effectiveness of climb(s). Basically cycle time.

Match the above with what your team needs.

microbuns · January 8, 2018, 4:15pm

OPR isn’t great in this game, but I think CCWM will be a good predictor.

Anthony_Galea · January 8, 2018, 5:02pm

A theoretical robot that scores 1 cube a match in their own switch against a majority of their opponents’ ‘kit bots’ (no appendages) will have a higher OPR (if I’m not mistaken) than a good amount of teams at the championship who are very capable but have a difficult match schedule. I think match scouting is going to be crucial this season, recording where a team placed its cubes and how well they fared under defense. I would completely ignore OPR this season.

But that’s just my 2 cents.

Caleb_Sykes · January 8, 2018, 5:18pm

I anticipate that the predictive power of OPR this year will be comparable to that of OPR in 2006, 2007, and 2017, which were the worst years on record for OPR. There is even a reasonable chance that this will be the worst year on record for OPR. I think if there was ever a year for Elo to really prove its potential, this would probably be it. I’ll post an analysis after week 1 to see which has greater predictive power.

CalTran · January 8, 2018, 5:47pm

I suspect that this year, actual points scored by robots gets tossed out the window for scouting. Personally, I’d be interested in recording robot ability (like if and where they can score cubes), drive team ability to work with others, and mechanisms used to achieving Climbing end game, and all of this applied to compatibility with my own robot.

Paul_Richardson · January 8, 2018, 6:23pm

Here’s a post I wrote in another thread.

Scoring Theory for Power Up

I’m interested in seeing people’s ideas for stats that could be used like OPR normally would. Basically, points scored on Scale or Switch have almost no correlation with robot capability, so OPR isn’t useful.

DPR has never really been used much, but I think it could be rough rating of robot capability for this year, if you run it on only the Scale/Switch score data from FMS. If you cause other alliances to score less than they normally do, that means you’re probably placing cubes to do that. Since the scoring tells you fairly directly that one alliance is “better” than the other, but not by how much, you could mix in the components of an Elo system (an Elo system on its own would probably be pretty good, too) with successive solutions of the DPR score matrix after each match contributing to a running Elo rating total, weighted based on opponent DPR. I’m still trying to figure out exactly how the math for this could best be set up (definitely needs real match data and testing).

Depending on how much score data you get from FMS, you could possibly count how many times Ownership changes, which would help you rank teams relative to each other. If two teams change Ownership a lot, they are similarly capable. If one alliance has Ownership the whole match, then that alliance is clearly better. This also lends itself to Elo rating.

OPR might be useful in the context of climbing, if ramp/platform robots are widespread, or in cases where two robots cannot climb together because of conflicting designs. With those robots, you’d see robots who don’t normally climb suddenly getting those points, or vice versa. If you use a robot’s Climb OPR to find their expected climb contribution, you can compare that to their actual performance in each match. Do that for each of a robot’s partners throughout the competition and you figure out if that robot is a “facilitator” or a “blocker”.