Where to find explanation of calculations behind OPR?

FIRST General Forum
1

Hey CD… long time no see.

I have a challenge in front of me at work that, I think, might be solved using a similar approach to how FRC teams calculate OPR. I understand the high-level idea behind those calculations, but haven’t done a deep dive into them.

Does anyone happen to know of a resource I can read that will walk through how the equations are derived and solved?

Thanks. Hope everyone is excited for another year of FRC. This is going to be my 20th, somehow. Time flies.

2

I found this presentation to be very helpful:
https://www.chiefdelphi.com/media/papers/download/4025

3

A quick search turned up this thread.

Ether (the resident engineer here on CD and an absolutely brilliant mind) seemed to spearhead that discussion. I’m currently reading through it for the second time and learning plenty.

The basis is this: a big set of equations that must be simultaneously solved to determine a teams average contribution to an alliance. Lots of matrices and good stuff like that.

4

You may want to check out Mini-tab and its regression analysis tool. I used it for work many years ago at calculating the cost of different features in a product. I had several similar products (cylinder heads) with slightly different feature content. Minitab was able to help me predict the average cost per feature and what reductions should reflect in terms of unit pricing.

5

As IKE mentioned, OPR is calculated using linear regression. You can google basic equations to do so, or use different graphing softwares. I know that tableu also has the function built in.

However, I would not suggest using OPR in scouting statistics. I think that looking at individual team averages(from competition match scouting), is much more beneficial. Our team currently uses google sheets for visualization and analytics.

6

OPR is calculated using multiple linear regression.

You can google basic equations to do so, or use different graphing softwares. I know that tableu also has the function built in.

I’d be very surprised if Tableau supports multiple linear regression sufficiently optimized to handle large datasets (like World OPR).

7

*To do large-scale OPR-style computations (like World OPR), a stats program must:

  1. support multiple linear regression (most do), and

  2. have an option to suppress the constant parameter (many don’t), and

  3. be capable of handling large datasets (e.g. 18,000 equations in 3,000 variables), and

  4. be optimized to handle such large datasets (i.e. use sparse matrix technology)

There are very few free stats programs that meet all the above criteria.

There are free numerical computation packages that meet all the above criteria.