Newton 09

[1t5h1e1o]

Originally Posted by artdutra04

Now that would be a sight to see - a Norm Abrams FRC robot!

That thing would have some major class.

[Joe Ross]

Originally Posted by TKM.368

Thanks Joe! I don't suppose you intend to rerun the simulation with the addition of 687, will you?

No promises. It takes about 24 hours to run a simulation (most of that is generating matches). I need to refactor the code to add another team, and I'm not sure if I'll have time to do that and run the simulation before the championship.

687's average OPR is similar to 33 and 768, so they should do similarly.

They won the engineering inspiration award at the Los Angeles Regional, which is how they qualified. Not sure how they added so late.

[BradMello]

Great show of teams in Newton this year, I look forward to the competition.



http://www.youtube.com/watch?v=1eov0n4pukE

total coincidence our best score was at 1:21

[Dan 1038]

Originally Posted by Joe Ross

Similar to what I did last year, I ran through 193 simulated qualification match schedules (all 102 matches, using the FIRST match generator). I used each team's average opr for their contribution to a match. I calculated the standings for each of the 193 simulated qualification schedules, and then looked at some statistics from the aggregate results.

When I went back and looked at last year's data, a team finished within 1 standard deviation of their average seed 60% of the time, and within 2 standard deviations 95% of the time. Only 3 teams seeded higher/lower then their min/max seed.

Team  Avg Seed	Median	Mode	StDev	Min	Max	#1 seed	Top 8
1625	6.3	4	1	7.5	1	48	40	155
121	7.4	5	1	7.7	1	47	46	131
1726	10.4	7	1	9.8	1	47	20	106
1507	12.6	9	3	12.6	1	69	11	96
2970	14.4	11	4	12.7	1	67	10	85
1701	17.6	14	3	13.6	1	78	7	56
234	18.5	14	5	15.3	1	65	7	60
16	21.8	17	7	16.2	1	76	4	47
1155	21.5	19	5	14.9	1	66	3	42
2377	22.4	19	5	15.8	1	76	2	44
102	21.2	17	7	16.1	1	67	3	49
191	21.9	19	2	14.8	1	64	2	43
364	22.6	19	31	15.5	1	67	2	40
1038	22.9	18	7	16.8	1	79	2	41

First, I must say I am impressed at your analysis - neat way to look at the available data and extrapolate a result. Unfortunately, as the stats suggest, the analysis really doesn't mean anything. The standard deviations are too large to lead any credance to the results, as would be expected in an analysis with this many variables. Keep at it, if you can refine this it will be a very cool way to predict team's success!

[LadyinthePit]

Originally Posted by David Brinza

Wait, Jim. I didn't see a miter saw, planer, router or jointer listed for 1730's pit. Or do you bring your own??

Let's put it this way at the Palmetto Regional Bilfred said that our team had everything but the kitchen sink. So you'll just have to come by and see what tricks we have up our sleeves

[Joe Ross]

Originally Posted by Dan 1038

First, I must say I am impressed at your analysis - neat way to look at the available data and extrapolate a result. Unfortunately, as the stats suggest, the analysis really doesn't mean anything. The standard deviations are too large to lead any credance to the results, as would be expected in an analysis with this many variables. Keep at it, if you can refine this it will be a very cool way to predict team's success!

It worked better last year because there was a much wider distribution of scores. This year, there's isn't as big of difference between the 3rd or 4th best robot in a division and the 20th. The standard deviations last year were about half of what they are this year.

The other thing this demonstrates is how only 7 seeding rounds isn't adequate.

The final factor is that this year it's much harder to separate a single robot's performance in to any single number. For example, your partner that pins another robot so you can score doesn't score anything itself, but certainly contributes points to the alliance. Last year's game was much more seperable, and in that case OPR correlated with our scouting very well (0.9). This year it's still decent, but not as good (0.6).

[BandChick]

For anyone interested I posted some fun signature userbars over here:
http://www.chiefdelphi.com/forums/sh...348#post849348

Support Newton! Or, if you can't pick, there's even one for every division!
FTC bars too, in case your team is competing in both!

[mark johnson]

[quote=BradMello;849216]Great show of teams in Newton this year, I look forward to the competition.




I agree,with a list of teams kike this 16 33 85 121 126 135 148 177 191 233 234 340 365 469 1625 1726 1732 1918 and im sure I missed some great teams ,Newton is rock solid!!!!!!!!!

[Dan 1038]

Originally Posted by Joe Ross

The final factor is that this year it's much harder to separate a single robot's performance in to any single number. For example, your partner that pins another robot so you can score doesn't score anything itself, but certainly contributes points to the alliance. Last year's game was much more seperable, and in that case OPR correlated with our scouting very well (0.9). This year it's still decent, but not as good (0.6).

Agreed, Last year was perfect for this type of analysis as there was no significant defense contribution, as Simbotics so blatantly showed all of us . We had the (mis?)fortune of being against them at both Midwest and the championships... This year has so many variables, it is about impossible to analyze individual robots using alliance stats. At Buckeye, I saw a number of defensive bots with no scoring ability score huge - they had human players off the basketball team! That is a key to the game this year which is kind of new, since the HPs can make or break a close match. It is also about impossible to show the effectiveness of HPs without dedicating scouts to watch the robot and another to watch the HP!

[Wayne TenBrink]

All these predictions and rating systems are great, but the only score that matters is the one that shows up on the board after the match. All are good in that they make us think about what factors matter in this game, and they give us something to do while we wait for the real thing!

All of them seem to be in general agreement, with the same teams showing up in approximately the same place on each. I would caution against trying to get too precise with any of this. In fact, I think you are kidding youself if you have more than one or two significant digits in the rating.

We plan to use advance scouting for two things:
1) Qual match planning - who are the scoring threats, who is likely to come after us, and who should we send after their top threat?
2) Draft scouting priority - since we really can't do a good job of scouting every team. We plan to focus on about the top half.

Not to be left out of the fun, we came up with our own system. It is based mostly on the FIRST in MI system (win/loss record, draft selection, and elimination performance, all normalized to account for the different numbers of events) with a scaling factor based on the Simbots "+/- contribution" thrown in for good measure. Did we put the proper empahsis on the correct statistics? Who knows. It must be right because it puts us in the top 10 (just kidding). It agrees with a lot of the other ratings, but there are some good teams that didn't make the cut, so I wouldn't bet the farm on it. Here is what we came up with for the teams above the median:

Team # Rating
1625 49
2970 43
1155 42
121 42
1038 41
1726 41
1657 39
368 39
1507 38
85 35
1918 34
852 33
234 33
612 32
1086 32
3075 31
16 31
2344 31
832 29
148 28
469 28
1868 28
1569 27
292 27
768 27
365 27
1706 26
126 26
2866 25
2377 25
135 25
2004 25
102 24
2996 24
2609 23
2659 23
1714 22
1629 22
360 22
88 22
1511 21
233 21
364 20
1732 20
1701 20

[rees2001]

Not sure how statistic driven your analysis is. You seem to have left off a good number of good teams and included a number of teams that were picked teams at regionals, not the pickers.

[Athleticgirl389]

Originally Posted by Joe Ross

Similar to what I did last year, I ran through 193 simulated qualification match schedules (all 102 matches, using the FIRST match generator). I used each team's average opr for their contribution to a match. I calculated the standings for each of the 193 simulated qualification schedules, and then looked at some statistics from the aggregate results.

When I went back and looked at last year's data, a team finished within 1 standard deviation of their average seed 60% of the time, and within 2 standard deviations 95% of the time. Only 3 teams seeded higher/lower then their min/max seed.

Team  Avg Seed	Median	Mode	StDev	Min	Max	#1 seed	Top 8
1625	6.3	4	1	7.5	1	48	40	155
121	7.4	5	1	7.7	1	47	46	131
1726	10.4	7	1	9.8	1	47	20	106
1507	12.6	9	3	12.6	1	69	11	96
2970	14.4	11	4	12.7	1	67	10	85
1701	17.6	14	3	13.6	1	78	7	56
234	18.5	14	5	15.3	1	65	7	60
16	21.8	17	7	16.2	1	76	4	47
1155	21.5	19	5	14.9	1	66	3	42
2377	22.4	19	5	15.8	1	76	2	44
102	21.2	17	7	16.1	1	67	3	49
191	21.9	19	2	14.8	1	64	2	43
364	22.6	19	31	15.5	1	67	2	40
1038	22.9	18	7	16.8	1	79	2	41
469	22.2	19	4	15.8	1	70	2	42
148	24.3	20	13	16.3	1	82	2	24
1732	25.4	22	19	16.1	1	68	1	27
135	24.2	22	30	15.2	1	77	6	31
1918	27.6	25	32	16.1	1	75	2	19
368	29.6	27	18	18.0	1	86	3	18
2344	28.7	27	6	17.3	1	86	3	26
1868	29.0	25	17	17.0	1	82	1	22
233	29.9	28	11	18.8	1	81	4	24
292	32.4	30	9	18.6	2	79	0	14
612	31.6	31	44	17.9	2	78	0	22
1714	33.4	31	22	17.9	3	85	0	11
852	32.5	30	8	19.2	1	87	1	17
360	31.2	27	51	19.2	1	78	2	22
85	30.9	28	25	18.3	3	84	0	15
126	30.4	29	22	16.4	1	77	1	14
122	35.9	34	43	18.9	1	84	1	13
1657	33.5	31	21	19.6	1	80	1	20
1629	33.8	32	32	19.2	2	86	0	17
88	37.8	35	19	20.8	2	86	0	15
846	35.1	34	30	18.7	1	83	1	13
365	36.7	34	14	20.6	2	83	0	12
768	37.9	36	35	17.4	2	79	0	7
33	36.9	36	32	19.2	1	81	2	11
1086	40.2	40	43	20.4	2	80	0	12
159	40.5	38	39	19.3	2	82	0	7
340	42.4	43	65	19.6	2	85	0	8
842	43.8	42	52	20.0	2	84	0	6
1350	43.9	45	46	18.9	2	82	0	4
3075	43.8	43	34	19.8	4	86	0	9
862	42.4	44	49	19.0	2	81	0	6
2890	44.1	46	55	19.9	2	85	0	9
2609	44.3	44	44	20.5	5	87	0	4
2866	46.0	47	34	19.1	6	87	0	4
177	46.6	51	53	20.7	6	84	0	3
2875	49.5	50	70	18.9	6	86	0	3
2283	51.7	53	68	18.0	10	83	0	0
2771	50.5	51	51	19.4	3	87	0	2
832	54.9	57	53	19.4	8	87	0	1
195	54.3	57	58	17.9	2	87	0	1
1138	54.3	57	61	17.9	12	87	0	0
578	53.4	56	56	18.0	6	86	0	1
1557	57.6	59	56	16.9	17	85	0	0
1730	53.9	55	38	20.0	6	86	0	2
228	56.3	58	71	18.7	1	87	1	2
1811	58.4	60	62	17.7	7	87	0	1
339	56.1	59	67	18.6	8	87	0	1
1516	57.2	59	55	17.8	10	87	0	0
1706	57.2	60	69	18.3	14	87	0	0
1098	56.8	58	72	19.1	7	87	0	1
358	59.3	61	70	18.5	7	87	0	1
2659	57.3	60	60	19.7	4	87	0	1
1569	60.1	62	73	16.8	12	85	0	0
980	58.5	60	66	17.7	11	87	0	0
2004	60.6	65	79	19.7	10	87	0	0
1511	60.2	64	58	18.4	8	86	0	1
1209	62.3	65	80	17.4	10	86	0	0
1458	61.9	67	56	17.6	12	87	0	0
2836	63.4	68	73	16.7	6	87	0	2
620	63.8	64	82	15.9	8	87	0	1
1506	67.3	72	75	16.2	25	87	0	0
2067	67.9	72	72	15.4	15	87	0	0
2702	67.5	72	81	15.6	11	87	0	0
2783	67.6	71	85	14.6	25	87	0	0
138	71.3	75	85	13.7	21	87	0	0
1547	69.4	73	83	14.0	16	87	0	0
1700	70.3	74	84	13.6	34	87	0	0
1311	70.4	74	74	13.7	25	87	0	0
1023	72.3	76	86	13.3	29	87	0	0
2996	73.2	76	84	12.0	22	87	0	0
2549	73.1	76	87	12.9	14	87	0	0
529	75.2	78	86	11.4	38	87	0	0
86	82.4	85	87	7.4	42	87	0	0

Boy oh boy do I love Stats <3 LOL! Coinicidence.... all 102 matches and 102 is there, doing pretty well... I always knew I liked that number lol. Seeing Stats. in general are just fun things so this is amazing

[Chris is me]

Originally Posted by Wayne TenBrink

All these predictions and rating systems are great, but the only score that matters is the one that shows up on the board after the match.

That's true, but you can't use that score to scout individual teams! There are 5 other robots on the field, that all affect the score! Making a scouting system based on that not only defeats the purpose of scouting at a regional, it's inaccurate because qualifiers aren't perfect.

Maybe I'm just critical because my team's not on the upper half of that list, though

[IKE]

Great work with all of the techniques so far.

One thing I like about the Simbots system is they look at each indiviual team and event. One important factor on this is to look for trends. Like anything else with time, some get better. Some do worse (because the field got better), some got lucky, some are consistently good.

The averaging systems can be dangerous because they do not pay attention to these trends. Some experienced teams had a great first weekend because there were a lot of dead bots at the regionals they attended that they could pick on.

Consistent top and bottom performers are easy to pick out. If a team only attends 1 event, it is very difficult to determine Lucky vs. Good.

Happy Data Crunching!

Rembmer always temper your data with common sense, and make sure your opinion reflects reality by using a data.

[rees2001]

Originally Posted by IKE

One important factor on this is to look for trends. Like anything else with time, some get better.

The averaging systems can be dangerous because they do not pay attention to these trends.

This is a good point, some teams show up for competition with systems that need some work while others show up as ready as they will ever be.

340 showed up with some work to do. We started 0-3 & then went 9-1 before losing 2 in the finals. The record shows 9-6 but I see, streaks of 0-3 then 5-0 in quals, 4-1 then 0-2 in elims. We have also had weeks to work out the bugs & make some changes . I love the speculation leading up to Championship & I do know that prior performance plays a part but, things can change.