Easy to use Offensive Power Rankings (OPR) program for mid-regional scouting - Page 7

SteveGPage · #91 13-03-2009, 12:16

Quote:

Originally Posted by Killraine

So in time for Chesapeake?

Yes, that is my plan! If I can't get a release on CD by then, I plan on having a working copy on my machine. Just come and find me in the stands on Friday and I'll be happy to either share the app, or at least be able to share the results with you. I try and sit in the upper section of the stands - to the right of the field (near the air exchange vents - which has an outlet to plug into!)

Steve

Joe Ross · #92 14-03-2009, 09:30

With v6, I was not able to get predictions for the Los Angeles regional (CA). Am I doing something wrong, or did is something on FIRST's side screwy?

Bongle · #93 14-03-2009, 10:11

Quote:

Originally Posted by Joe Ross

With v6, I was not able to get predictions for the Los Angeles regional (CA). Am I doing something wrong, or did is something on FIRST's side screwy?

Sorry, v6 has a bug with predictions. I'll upload a fixed version soon.

Edit: v7 now up. The only external improvement (I think) is that predictions are fixed and now work mid-way through a regional.
Edit2: Fixed a parsing issue with single-digit teams.

The Lucas · #94 14-03-2009, 11:49

Quote:

Originally Posted by Bongle

Sorry, v6 has a bug with predictions. I'll upload a fixed version soon.

Edit: v7 now up. The only external improvement (I think) is that predictions are fixed and now work mid-way through a regional.
Edit2: Fixed a parsing issue with single-digit teams.

Can you release the source as well?

I m looking into giving it a run offline option (-o) in case there is no internet publically available at the Philly regional. I am thinking about saving all the html results files in a local directory structure (instead of the temp file). That way on Sat I can update the html file by hand and keep re running the calc. I am not sure if I will have time to get this working or not

Thanks,
Brian

Bongle · #95 14-03-2009, 11:52

Quote:

Originally Posted by The Lucas

Can you release the source as well?

I m looking into giving it a run offline option (-o) in case there is no internet publically available at the Philly regional. I am think about saving all the html results files in a local directory structure (instead of the temp file). That way on Sat I can update the html file by hand and keep re running the calc. I am not sure if I will have time to get this working or not

Thanks,
Brin

See attached.

SteveGPage · #96 15-03-2009, 11:14

Quote:

Originally Posted by SteveGPage

Yes, that is my plan! If I can't get a release on CD by then, I plan on having a working copy on my machine. Just come and find me in the stands on Friday and I'll be happy to either share the app, or at least be able to share the results with you. I try and sit in the upper section of the stands - to the right of the field (near the air exchange vents - which has an outlet to plug into!)

Steve

We were able to finally get OPR+ and DPR+ values calculated from the equation that Bongle and I had both posted. Attached is the perl script that my son and I were able to get together this weekend.

Usage: cmd line: perl scouteval.pl {qualmatch}.txt ---- which will create a file called results.txt, and where {qualmatch} is the file created from the usfirst match results page, copied into a text file. Then use whatever software you use to solve the matrix for the team OPR+ and DPR+ values.

The OPR+ and DPR+ values (sorted by OPR+), as calculated for the DC regional, is as follows:

Code:

Team Number	OPR+	DPR+
118	78	38
2199	72	6
116	68	44
1731	66	32
1712	66	26
176	64	8
2068	62	32
1629	60	28
836	60	18
2961	58	26
2964	56	50
181	56	34
597	54	40
122	54	22
1418	54	10
45	52	54
1872	52	30
449	52	20
401	50	36
365	48	0
3046	46	34
1446	46	12
1915	44	36
272	42	34
339	42	26
587	42	26
611	42	12
2913	40	28
709	40	26
2914	40	16
53	40	16
1900	36	32
1748	36	8
620	34	16
1123	32	26
357	32	24
2819	30	40
1522	30	12
234	30	6
2911	30	0
346	28	8
538	26	18
2377	26	10
615	24	22
1719	22	46
768	22	26
1793	22	22
2962	20	12
7	20	12
2421	20	-4
1849	18	6
623	18	-10
614	18	-14
1111	16	6
2963	14	2
1727	12	28
2912	12	4
2900	10	2
1370	10	0
2121	8	-2
1885	8	-6
1279	6	26
2729	6	-4
2537	2	34
1699	-16	16

A couple of things I noticed, if you run the OPR+/DPR+ numbers with incomplete data, the numbers will look wrong, but relative sorting will be pretty close - but expect numbers like +800 to - 800. When all of the matches finish, the numbers should settle out to values like above. Also, we noticed that we could predict the outcome of the elimination matches about 65-70% of the time.

We are working to automate the process.

Let me know what you think!

Steve

martin417 · #97 16-03-2009, 11:03

I get this error when I run Oprnet on some regionals. (BAE for one) I know Peachtree and SanDiego work, but haven't found others that do yet.

Quote:

Downloaded 34342 bytes
File download complete
Parsing!
No matches found. This regional may not have run yet, or may have HTML output that the parser does not recognize.
Failure to parse XML. Code: -2147467259

Bongle · #98 16-03-2009, 11:33

Quote:

Originally Posted by martin417

I get this error when I run Oprnet on some regionals. (BAE for one) I know Peachtree and SanDiego work, but haven't found others that do yet.

Look at the FIRST-provided results for BAE. They don't exist for the qualification matches, and so OPRNet cannot run. Some people in this thread have written parsers that work with TBA results, which are posted. Check the whole thread and see if there is an app posted that can read TBA results.

JesseK · #99 16-03-2009, 13:13

Quote:

Originally Posted by Bongle

Look at the FIRST-provided results for BAE. They don't exist for the qualification matches, and so OPRNet cannot run. Some people in this thread have written parsers that work with TBA results, which are posted. Check the whole thread and see if there is an app posted that can read TBA results.

Part of the reason I extended my program to pull from multiple sources was to eventually put all of the data into a standardized, formatted .csv file that any program or Excel sheet could read. That way, we could pass around a master file with our programs for those who do not have the internet immediately available (or are blocked by a firewall for whatever reasons).

The format for a line will be something like

REG_CODE,MATCH_NUM,RED_1,RED_2,RED_3,BLUE_1,BLUE_2 ,BLUE_3,R_SCORE,B_SCORE

Where REG_CODE is the usfirst.org regional code. There will be 1 line for every match at every regional (roughly 48*76 = 3648 lines). Once the matches are verified I'm also going to put in a new 'feature' that writes the parsed lines directly into the Java source code for the next time the code is compiled (which is often for me since I use a similar program to do my Fantasy FIRST picks). The order it will check for valid data is source code --> local parsed file --> local html file --> usfirst.org --> tba.net ... and it will go until it finds valid data. Buckeye data still isn't up though, so I wonder what we'll have to do to get those scores

Then, eventually I'll put a scouting layer on top of that so you can input data into it at a regional. Then, I'll make a simple version of this input, so someone can input it via PDA, iPhone, or whatever other touchpad technology comes available. This part won't happen publicly till the offseason though.

I've been away from it for a few days since I was at the FL regional, but now it's back to work to have it done in time for Atlanta.

-- edit -- Ooh and a w00t to Greg Marra and crew for simplifying the tba.net urls for each regional and year!

martin417 · #**100** 16-03-2009, 17:58

I ran through all the regionals that worked (three didn't) and put each one on a spreadsheet. I also combined tham all into one sheet (master). hope this is useful to teams.

Lil' Lavery · #**101** 16-03-2009, 18:09

Quote:

Originally Posted by SteveGPage

We were able to finally get OPR+ and DPR+ values calculated from the equation that Bongle and I had both posted. Attached is the perl script that my son and I were able to get together this weekend.

Usage: cmd line: perl scouteval.pl {qualmatch}.txt ---- which will create a file called results.txt, and where {qualmatch} is the file created from the usfirst match results page, copied into a text file. Then use whatever software you use to solve the matrix for the team OPR+ and DPR+ values.

Spoiler for length:

I love sabremetrics, OPR/DPR, +/-, and stats in general. I was excited for these OPR+ and DPR+ numbers. But unfortunately, just from looking at them, I can tell we might have taken it a step too far and they might have become less accurate.
The fact that my former team, 116 is ranked 3rd at the event should be the first indicator (as well as the ranks of 45, 365, and 234 being so low), despite struggling to score the entire regional. I don't have the raw OPR numbers handy (on my Mac), so I can't give detailed comparisons between the rankings, but judging by these, I think it may have become less accurate to what the team's real performance was.

SteveGPage · #**102** 16-03-2009, 21:42

Quote:

Originally Posted by Lil' Lavery

I love sabremetrics, OPR/DPR, +/-, and stats in general. I was excited for these OPR+ and DPR+ numbers. But unfortunately, just from looking at them, I can tell we might have taken it a step too far and they might have become less accurate.
The fact that my former team, 116 is ranked 3rd at the event should be the first indicator (as well as the ranks of 45, 365, and 234 being so low), despite struggling to score the entire regional. I don't have the raw OPR numbers handy (on my Mac), so I can't give detailed comparisons between the rankings, but judging by these, I think it may have become less accurate to what the team's real performance was.

I have similar concerns. Last night I ran every regional that we had numbers for. I had hoped to create a combined "top 25" list from every regional, until I saw which teams rose to the top, and which teams did not. I did a double check on TBA to see how these teams had actually performed, and was surprised that teams in some regionals - such as WI, NYC, and Oregon - were showing up on the top despite the fact that many had only moderate records, and did not appear to be high scorers. Some regionals, the numbers corresponded with what I expected, but other regionals, the numbers did not. The one factor that I'm still not sure about, and perhaps you can help me with this Sean - the team scores are not only the robot - so the purpose of this equation was designed to look at not only how effective the robot is, but also how well the HP does, too.

During DC, I had my scouts do extensive scouting on each match, looking at how many MRs each scored, and how many MRs were scored on them (developing a match by match +/-). What I failed to do, was to have them track the HP and how many SCs the team delivered to the HP, both of which had profound impact on the game. If I just looked at the performance of each robot, 45, 365, and 234 were on the top of my list. But I knew from observations that 2199's HP was deadly, and 118 had the ability to get at least 2 SCs exchanged. So I wonder if there isn't some aspect of the OPR+ calculation relying on those other factors, and not just how effective the robot is.

I'm going to run these numbers (looking at all three factors - robot, HP and SCs) while we are in the midst of the competition at Chesapeake this weekend to see if I see any trends. Perhaps this equation isn't going to give us the info we are looking for, but I still want to run some additional numbers to see how they do.

Best regards,

Steve

AlexD744 · #**103** 16-03-2009, 22:08

It was perfect because we were #1 after Friday. And therefore it must have been the best tool ever. Just Kidding, however, it really did help in our meeting. It confirmed that KRUNCH was as amazing as we thought.

SteveGPage · #**104** 16-03-2009, 22:45

Quote:

Originally Posted by AlexD744

It was perfect because we were #1 after Friday. And therefore it must have been the best tool ever. Just Kidding, however, it really did help in our meeting. It confirmed that KRUNCH was as amazing as we thought.

The numbers you are referring to, are the OPR/DPR numbers. The numbers that we are talking about are the OPR+/DPR+ numbers - which, yes, I know don't look a lot different!

I pulled the data from FL, and ran the calculation for the finals, using the OPR+ and DPR+ numbers. Here are the results ...
First, averaging the score across all three finals matches, the avg Red score was 74 and the avg Blue score was 84.
I plugged the calculated values into the OPR+ and DPR+ formulas, and here is the outcome Red 68, Blue 90. An 8% difference for Red and a 7% difference for Blue between the predicted values and actual values. Penalties, which are not considered in the calculations, could easily account for that kind of discrepancy.

In doing so, one thing became obvious, the OPR+ number can not be taken into consideration by itself, but must be linked with the DPR+ number - but not simply by adding them together - but in the separate calculations of Red vs Blue Score. I'm going to do some additional modelling of the data to see how we can get a combined ranking value.

I am going to run additional calculations at other regionals and see how they correspond.

Steve

andrew418 · #**105** 17-03-2009, 16:10

Hi, I've been trying to access the ratings for the Dallas regional, and the program has not been able to parse the html data. Does anyone know why?

View Poll Results: Was this useful?
Yes, it was! It helped point out diamonds in the rough	109	70.32%
No, its numbers generally did not correspond to robot's actual on-field performance	46	29.68%
Voters: 155. You may not vote on this poll

Thread Tools
Show Printable Version Email this Page
Display Modes	Rate This Thread
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode	Rate This Thread:

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Offensive Power Ranking Calculations	Greg Marra	General Forum	4	18-03-2009 22:03
Offensive Power Rankings for 2008	Bongle	General Forum	166	18-05-2008 15:05
Offensive+Defensive Power Rankings	BornaE	General Forum	4	05-04-2008 01:33
2006 Offensive Power Ratings	sw293	General Forum	16	10-05-2006 17:04
(CURIE) Let's make this Easy for all of us. [scouting]	157#1Driver	Championship Event	13	22-04-2002 08:56