Re: OPR Formula
 

Phil, can you explain how you got the info from twitter?

Re: OPR Formula
 

The problem with DPR is that when teams do not play defence, DPR can be unpredictable. While it will take into account offensive hoarding by the team (if applicable), this is not always significant compared to mostly-random fluctuations, including schedule.

Occasionally, DPR will take into account non-random factors other than the team's robot, though. In 2011, a team's human player could play a significant role in determining opponents score. A human player's output (# of tubes thrown) and accuracy (% of tubes picked up by own alliance) definitely affect opponent score, for example. A human player who threw many tubes or wasn't very accurate could result in a team having a high DPR.

In 2012, human players aren't a factor, but hoarding as an alliance strategy certainly is (note that an alliance could control 15 of the 18 balls at a time). A low DPR by a team not playing defence would probably be indicative of such a strategy.

Re: OPR Formula
 

I didn't get it from twitter. I got it from the web based results (with a URL like this: http://www2.usfirst.org/2011comp/eve...chresults.html) by just going through all the event codes and putting it as a parameter in an excel query that would get the data from http://www2.usfirst.org/2011comp/events/event_code/matchresults.htm. Then, I just put them all in the table I posted.

To get this data from twitter, you'd have to use the API, and I'm not sure to go about that. Maybe you could fetch tweets using JSON?

Re: OPR Formula
 

Hey guys,

I took what all of you have contributed (great post by the way, very helpful) and constructed an excel file to do RUSH's post-competition analysis, or other competition scouting for states and nationals. So i wrote a macro (not very efficient at those, sorry) and it creates webquery files, obtains the data, and organizes it for however many teams are in the tournament. The only problem I am having is the OPR calculations. Can someone look at my macro and see what they can do about adding an OPR calculation formula or something? Or give me some feedback on how i can improve it cause I really am not too good at these. Thanks.

Re: OPR Formula
 

That's along the same lines I was thinking. If you could calculate the opposing score, then you would have your DPR.

Re: OPR Formula
 

Stingray, nice work! I'm going to include that in our OPR this year so we can auto-update it on the fly. Really outstanding.

Re: OPR Formula
 

Hi,

Nice work. If you want your spreadsheet to calculate OPR from the results that you get from the web query, you are welcome to copy any of the macros that I published. I did not put protection on them to allow other teams to use them and customize for their own use. However if that is all you are going to do, I don't see the purpose of it except as a learning process since I publish OPR/CCWM results of every regional and district after each week of competition and I have been doing it for the last 4 years. Please refer to the following white paper.
http://www.chiefdelphi.com/media/papers/2174
You will not see the weekly published spreadsheets from prior years since I delete them at the end of the season and only keep the last one.

Re: OPR Formula
 

Attachment shows OPR, CCWM, & DPR scores obtained by solving the 2052 simultaneous equations obtained for the 2052 teams in that "one big file".

On my 8-year-old computer, it took ½ second to read the file and create the 2052x2052 matrix. It took 19 seconds to factor the matrix and compute OPR, CCWM, and DPR.

Re: OPR Formula
 

And just for kicks attached is a histogram of the CCWM scores. I find it interesting how normal in shape the distribution is; I would have expected it to have a longer tail on the right than is apparent.

Re: OPR Formula
 

Hi Ether,

The numbers you have is slightly different than mine. Do you include elimination round matches?

Your 8-year-old computer must have been quite a powerful computer 8 years ago. Or is there a solver you used that is faster than the Cholesky Decomposition that I used. Here is the statistics that my son got when he was a junior in his independent study math class.

Computer Model Hewlett-Packard HP ProBook 6555b
Platform Mobile
OS Version Microsoft Windows XP Professional
CPU AMD Turion(tm) II P520 Dual-Core Processor
Memory 2807 MB

Here are the data table from tests: (time in seconds)
Size Gauss-Jordan method LU Factorization Cholesky Decomposition
100 0 0 0
300 9 1 0
500 46 8 1
750 159 27 4
1000 383 64 11
1400 1056 178 30
1800 2278 384 63

This was the 2010 data and it took 63 seconds on that slow machine.

Re: OPR Formula
 

Hi Ed,

The numbers you have is slightly different than mine. Do you include elimination round matches?

I used all the data in the "one big file" that Phil attached to post#8 in this thread. That file contains only raw data (no metadata) so I can't say for certain what Phil included.

@Phil: Can you jump in here and answer Ed's question?

Your 8-year-old computer must have been quite a powerful computer 8 years ago.

Pentium D 3.4GHz 1GB

Or is there a solver you used that is faster than the Cholesky Decomposition that I used.

I also used Cholesky, but all Choleskys are not created equal. There are various implementations. Some are column-oriented and some are row-oriented; some are in-place and some are not. It makes a difference. The one I used is a slightly modified version of one that I selected sometime back in the late 80s after testing several different algorithms from various texts. It's optimized for memory access. The 19 seconds was for a 2052x2052 matrix using double precision (64 bit) floats for the matrix and Intel extended precision 80-bit floats for intermediate calculations.

I just re-ran it with single-precision (32 bit) floats and it took a little less than 12 seconds to factor the 2052x2052 matrix.

1800 2278 384 63
This was the 2010 data and it took 63 seconds on that slow machine.

That's 63 seconds for an 1800x1800 matrix. Since Cholesky goes as O³, that would take ~93 seconds for a 2052x2052 matrix.

How are you computing the results in the "Worldrank" tab in the Team_2834 2011_Scouting_Database Championship v4 spreadsheet? My version of Excel (2000) only has 256 columns - not enough to hold a 2052x2052 matrix. I assume you are crunching the numbers in some other app?

I also computed the OPR & CCWM for the data (qualification matches only) in the link that Tom Line included in his post#4 in this thread. Attached are my results. The columns are Team#, OPR, CCWM, and DPR. If you would run your computation on the same data we could compare apples to apples.

Re: OPR Formula
 

You know, I might have actually forgot to put elimination rounds into that file. I forgot that I had to specifically tell excel to import a second table. Attached is a revised version of my entire workbook. Sorry about that, guys.

EDIT:
I attached another file that has some more information - regional names, match times, etc

Re: OPR Formula
 

Understand that in the twitter feed there is a lot of garbage data too. Various test posts as well as practice matches.

Re: OPR Formula
 

Attached is analysis for the data in the tab named "All Matches", less lines 5526, 5766, 6004, and 6196 (DQ).

Columns are Team#, OPR, CCWM, and DPR

Re: OPR Formula
 

Ether,

Your republished numbers are very close to mine now. I also used double precision. My implementation is done in Excel using VBA. I do not actually put the 2053 X 2053 matrix inside the spreadsheet so it works in older version. The program read in the match results and assemble the matrix. It is just stored in the memory for the calculation. Solving the equations inside Excel using VBA is probably slower than outside in another app. Tom Line mentioned they are solving the equations using Excel built-in solver. I am curious what algorithm they use and how fast that would be.

How come you only have 2052 teams? I have 2053 teams in 2011.