Week 4 OPR

[Ed Law]

Originally Posted by Navid Shafa

Since a comparative OPR isn't really available on the global scale at the moment, I was curious to see how we ranked amongst other teams.

Your team is ranked 29 out of 1548 teams who have competed so far. That is very good. You are one of the better teams. This ranking is what I called True World Ranking which is based on every qualifying and elimination match of every district/regional so far. All of that data is assembled into a giant matrix and solve so all interactions between teams are considered. Please see my old post here.
http://www.chiefdelphi.com/forums/sh...609#post964609

The data can be found here.
http://www.chiefdelphi.com/media/papers/2174

Navid, there are no OPR organizers here on CD. Nobody owns it. I didn't invent it. I think the person who did is no longer on this forum as I have not seen his posts in a long time. So don't worry about treading on somebody's feet. I am doing it as a service to others and it is my way to give back to the CD community who had helped our young team a lot in the last few years.

Nathan, thanks. You beat me to it, and I am not the only one staying up late.

[waialua359]

Ed,
did you include Hawaii?
I dont see it when I type in our team #.

-Glenn

[mwtidd]

Originally Posted by waialua359

Ed,
did you include Hawaii?
I dont see it when I type in our team #.

-Glenn

With ETC I have you at 52 (top for hawaii)
The most notable thing about your team is in qualifying matches, someone on your alliance got 1st place in every race.
With regards to the regionals I have full data on, you are the only team who has achieved this.

By pairing up with 368 you created a minibot monopoly.
This is another example where picking the best robot may not be the best pick.

It looks like 2439 had an great robot, but a less valuable minibot.
They picked 1056 who had the most valuable robot, and a decent minibot.

However after looking at 359 and 368's EMCs the only way to beat them would be to compete against them in the race. It would have been a gamle, but it looks like 2439 should have considered 2090, who looks like the only team with a minibot who could really go toe to toe with 359 and 368s.

Here is just another example where minibot monopolies win...
Congrats guys on the great job!

[Ed Law]

Originally Posted by waialua359

Ed,
did you include Hawaii?
I dont see it when I type in our team #.

-Glenn

Hi Glenn,

Thanks for catching the mistake. It has been updated in version 5.

http://www.chiefdelphi.com/media/papers/2174

I think what happened was when I first tried to get the data for all Week 4 events, your regional was not done yet due to time difference. I later went back to get the data, ran the True World Ranking macros and published the results and forgot to run the macros for the regional itself. So the True World Ranking was valid and did not change. The only difference in version 5 is that it includes the Hawaii regional results for the Query, CCWM results and OPR results tab.

Sorry about the error.

[mwtidd]

Originally Posted by Ed Law

Navid, there are no OPR organizers here on CD. Nobody owns it. I didn't invent it. I think the person who did is no longer on this forum as I have not seen his posts in a long time. So don't worry about treading on somebody's feet. I am doing it as a service to others and it is my way to give back to the CD community who had helped our young team a lot in the last few years.

Nathan, thanks. You beat me to it, and I am not the only one staying up late.

OPR has actually the product of collaborative work done by many individuals who are still on CD ( I know Jim Zondag worked on it, and he is still very active in the CD community)

OPR is standard matrix algebra and is a published mathematical formula, so when I see different OPRs it makes me a bit nervous. OPRnet may be taking across multiple event's rather than latest events, which would be cause for different values.

[Bongle]

Originally Posted by lineskier

OPR is standard matrix algebra and is a published mathematical formula, so when I see different OPRs it makes me a bit nervous. OPRnet may be taking across multiple event's rather than latest events, which would be cause for different values.

OPRNet just does the event that you specify. If you want just the most recent events, you'd have to figure out what their event codes are and enter them, one regional at a time.

Example: here's an excerpt from the bat file I use to generate the total list:

oprnet hi 2011 opr r q >> allranks.txt
oprnet is 2011 opr r q >> allranks.txt
oprnet dmn 2011 opr r q >> allranks.txt
oprnet nv 2011 opr r q >> allranks.txt
oprnet tx 2011 opr r q >> allranks.txt
oprnet ca 2011 opr r q >> allranks.txt
oprnet il 2011 opr r q >> allranks.txt
oprnet mn 2011 opr r q >> allranks.txt
oprnet mn2 2011 opr r q >> allranks.txt

Here's a list of every team at every regional so far.

[mwtidd]

Originally Posted by Bongle

OPRNet just does the event that you specify. If you want just the most recent events, you'd have to figure out what their event codes are and enter them, one regional at a time.

Example: here's an excerpt from the bat file I use to generate the total list:

oprnet hi 2011 opr r q >> allranks.txt
oprnet is 2011 opr r q >> allranks.txt
oprnet dmn 2011 opr r q >> allranks.txt
oprnet nv 2011 opr r q >> allranks.txt
oprnet tx 2011 opr r q >> allranks.txt
oprnet ca 2011 opr r q >> allranks.txt
oprnet il 2011 opr r q >> allranks.txt
oprnet mn 2011 opr r q >> allranks.txt
oprnet mn2 2011 opr r q >> allranks.txt

Here's a list of every team at every regional so far.

Oh okay! Thanks for clarify this, I have a mac so I can't use it.

[The Lucas]

Can someone please post Midwest OPR. The data up on the FIRST site has cannot be parsed because it is in a different format. It is probably due to the field issues they had during Elims. The same thing happened to Chesapeake and I posted the data.

[EDIT] Nevermind, Bongle fixed it in v14, here is IL

0	OPR	111	73.9248
1	OPR	118	51.1772
2	OPR	1625	45.6884
3	OPR	16	41.7728
4	OPR	2410	41.7174
5	OPR	1732	41.1471
6	OPR	135	37.2743
7	OPR	71	33.2081
8	OPR	2194	28.6006
9	OPR	2949	28.1996
10	OPR	2358	27.9021
11	OPR	2338	27.6187
12	OPR	2826	27.1633
13	OPR	1987	25.7962
14	OPR	2041	23.3367
15	OPR	1675	20.4011
16	OPR	3494	18.6332
17	OPR	45	16.518
18	OPR	2171	12.9152
19	OPR	1781	12.3256
20	OPR	2704	9.65644
21	OPR	2022	9.45619
22	OPR	3352	9.08313
23	OPR	2115	8.96639
24	OPR	101	8.44625
25	OPR	3779	8.22969
26	OPR	3595	6.91017
27	OPR	3067	6.79252
28	OPR	3197	6.37619
29	OPR	2151	4.5455
30	OPR	3612	4.44737
31	OPR	1739	4.27756
32	OPR	2432	4.04017
33	OPR	1367	3.10731
34	OPR	3488	2.54644
35	OPR	2769	2.35486
36	OPR	1091	1.68945
37	OPR	3646	1.40082
38	OPR	896	1.12364
39	OPR	3177	1.03324
40	OPR	3061	-0.270196
41	OPR	3695	-0.280014
42	OPR	2781	-2.29811
43	OPR	1850	-3.42507
44	OPR	2462	-3.43421
45	OPR	3110	-4.91922
46	OPR	3135	-7.98799
47	OPR	2803	-8.05401
48	OPR	2709	-8.28509
49	OPR	3416	-10.3558

[Bongle]

I just fixed the OPRNet parser to parse the "fat"-style regional results.

Congratulations to 111 for their performance this weekend!

[Nathan Streeter]

Originally Posted by lineskier

OPR is standard matrix algebra and is a published mathematical formula, so when I see different OPRs it makes me a bit nervous. OPRnet may be taking across multiple event's rather than latest events, which would be cause for different values.

Yes, it is disconcerting if people are getting different OPR numbers... But that really isn't the case: Ed's and Bongle's OPR's do match up for each robot for each tournament. The reason why Ed's OPR *World Rank* numbers are different than taking the highest OPR from each team is because it solves a single world-wide matrix. This is great in many ways, but has the (single?) downside of solving for a single OPR from their Week 1 performance to their Championship performance, which is somewhat flawed. That said, I'm not sure what'd be a better way... :-/

Lineskier (Mike?), would it be possible to use the matrices to compute the EMC and ERC? I really like that EMC and ERC use the twitter feeds to break down a Robot's contribution into Minibot and Hostbot, but without the matrices, they just don't have the same accuracy.

Thanks, EagleEngineer for pointing out the typo, I'll try to fix that!

[Bongle]

Originally Posted by Nathan Streeter

Lineskier (Mike?), would it be possible to use the matrices to compute the EMC and ERC? I really like that EMC and ERC use the twitter feeds to break down a Robot's contribution into Minibot and Hostbot, but without the matrices, they just don't have the same accuracy.

It seems like it should be possible to do this.

Let's say you had two variables per team:
OPRMini
OPRMain

Right now, the right-side vector of data for OPR is just a team's total scores because that's all I get out of the FRC match results. Since Lineskier has all the scores divided into points + bonus, he could just solve using minibot-only and main-only. The downside of this would be that you'd have teams with nonzero minibot scores who never had a minibot or never even went near the tower. You could say those teams "helped get their alliance minibot scored", but really it'd be mathematical noise.

[Ed Law]

Originally Posted by Nathan Streeter

Yes, it is disconcerting if people are getting different OPR numbers... But that really isn't the case: Ed's and Bongle's OPR's do match up for each robot for each tournament. The reason why Ed's OPR *World Rank* numbers are different than taking the highest OPR from each team is because it solves a single world-wide matrix. This is great in many ways, but has the (single?) downside of solving for a single OPR from their Week 1 performance to their Championship performance, which is somewhat flawed. That said, I'm not sure what'd be a better way... :-/

The World Rank is a season long overall score of how a team did. If a team went to two events and scored low on the first one and scored high on the second one, the single World Rank OPR will reflect that. It is not meant to show how strong a team is going into Championship. For that I would recommend using latest score which my spreadsheet will calculate in the OPR Results tab by putting -2 in the G2 cell and click the Calc box to run a simple macro.

Originally Posted by Nathan Streeter

Lineskier (Mike?), would it be possible to use the matrices to compute the EMC and ERC? I really like that EMC and ERC use the twitter feeds to break down a Robot's contribution into Minibot and Hostbot, but without the matrices, they just don't have the same accuracy.

Nathan, this has already been done by Team 33. But it was done manually as a one time thing. The data looks interesting. I was going to automate that using the twitter feed but when I heard that there are missing matches and may be even event, I changed my mind. I don't want to spend time on something if the data is not reliable. If anyone is interested in pursuing that, you are welcome to modify my macros. All the macros in my spreadsheet are not protected. It includes all code needed to assemble and solve the matrix for you.

[mwtidd]

Originally Posted by Ed Law

Nathan, this has already been done by Team 33. But it was done manually as a one time thing. The data looks interesting. I was going to automate that using the twitter feed but when I heard that there are missing matches and may be even event, I changed my mind. I don't want to spend time on something if the data is not reliable. If anyone is interested in pursuing that, you are welcome to modify my macros. All the macros in my spreadsheet are not protected. It includes all code needed to assemble and solve the matrix for you.

Nathan was asking about ETC, ERC and EMC which have nothing to do with matrices or OPR. And they are all calculated automatically. I just haven't published the latest version of my site.

Originally Posted by Nathan Streeter

Lineskier (Mike?), would it be possible to use the matrices to compute the EMC and ERC? I really like that EMC and ERC use the twitter feeds to break down a Robot's contribution into Minibot and Hostbot, but without the matrices, they just don't have the same accuracy.

Actually matrices inherently has an equal potential of inaccuracy. Digging through my threads I talk about how my algorithm is actually executed and the resolution it gets over OPR. Matrix algebra looks at the big picture, but it doesn't look match to match. ETC does look match to match. Point inflation affects both in different ways. Also I can tell which alliances finished in which place with minibots. So if someone has an EMC of 30 I KNOW in every qualifying match their alliance put up a winning minibot.

Also I calculate your team's contribution to each match, for the robot. OPR does as good a job as it can using traditional mathematics. I use CS algorithms to calculate my #'s. That being said there are still issues with point inflation that are inherent in any system. Some teams will have the luck of the draw. I have an idea of how to fix this for the ERC but it will take a bit to implement.

If someone had a java function for calculating OPR I could quickly implement it. ETC and OPR look at a teams contribution from 2 separate directions. I believe by utilizing both that the values it will hone in on a team's true contribution.

[IKE]

Originally Posted by Ed Law

Nathan, this has already been done by Team 33. But it was done manually as a one time thing. The data looks interesting. .

When compairing this data to some scouting data, it was reasonably accurrate. It turned into a bit of a "tube scoring" version of OPR. I have also eseen this done with penalty data, and some other interesting factors. We should have some good scouting data from Troy that hopefully we can use to tune the algorithms to be even more representative, and then test them at the State championship.

Like the robot, iterating and progressively improving your scouting and metric based scouting as the season continues helps a bunch.

[Ed Law]

Originally Posted by lineskier

OPR has actually the product of collaborative work done by many individuals who are still on CD ( I know Jim Zondag worked on it, and he is still very active in the CD community)

The origin of OPR was not a collaborative effort. I was talking about who coined the term and first proposed it. From what I can find, it was Scott Weingart from Team 293 in this post dated 4/6/2006, http://www.chiefdelphi.com/forums/sh...0&postcount=19. He is no longer active on CD as his last post was 4/13/2007.
I was not talking about who use it. Many people use it as a supplement to match scouting. Yes it is very simple matrix algebra and many have written their own programs to calculate it. If implemented correctly, the OPR numbers for each regional/district should be identical.

I am glad to see you proposing a new way of ranking teams. I always welcome innovation. However I do have serious concern about your method. This is not a personal attack so please don't be defensive. I just want to point out a few things so you can improve your algorithm if you choose to. I am not saying OPR is better. I am not defending OPR as I don't own it. In the end, you look at how numbers are calculated and you choose what you think will work for you.

1) One of my concern is if a team has a bad first match for whatever reason and that team's alliance score zero points. Then the next time this team play, according to your algorithm, you will assume that this team does not contribute much to whatever the score was for their second match. Please correct me if I misunderstood your algorithm.
2) I also read someone suggested to you that it should be iterative, i.e. loop back and have a second pass and third pass etc. I don't know if you tried this or not. An iterative method is fine as long as it converges. If it diverge or oscillate, then there is something wrong. In your method, does it always converge and if yes how many iterations typically before it converges. Have you compared the converged value to OPR? Are they close to each other?
3)One of the point you advocate about your method was considering match to match effect rather than a big picture like OPR. When you iterate, it is no longer just match to match. This is somewhat analogous to finite difference method. You are actually getting the effect of all the matches when you loop back and iterate.
4) If you argue that it should not iterate, then the final number is too dependent on your starting value and the method will not be mathematically valid.
I hope you will find a way to improve it so we will all benefit from a better way to rank teams.