White Paper Discuss: Karthik's Championship Event Data

Thread created automatically to discuss a document in the White Papers.

Karthik’s Championship Event Data by Karthik

I’ve taken all the data from the FIRST website from the regional season, and compiled it into one spread sheet. From there, I calculated a few stats and tried to make sense of it all.

For each regional here’s what I have.

Team #
Team Name (Hidden Column)
QP Avg. (0…2)
Seed
Seed % = (# of Teams + 1 - Seed)/(# of Teams)
Average Score (Just for qualifying matches)
Relative Average = Team Average / Regional Average
Standard Deviation
Factor: This is a stat that Rourke created, and he will talk about in more detail some time this weekend. The formula is:

(Team Avg. - Team Std. Dev.)/Regional Average

It’s pretty neat, as it rewards a team for scoring well, and consistency.

From there, I’ve created a metric which assigns values to Seeding, Averages, Championships, Finalists and Awards, to create a value to try and determine the “best” team at each event.

On the Master sheet, all the teams are placed together, so you can sort by any column to see the leaders in that category. You can also sort this sheet by championship division.

On the Master metric sheet, I’ve created a weighted average so each team has one score. It works like this; If a team went to one event, their score is their score from that event. If they went to two events, then I take the sum of two events and divide by 1.5. If they went to three events, I divide the sum by 2. The reason for this is, these stats are being used as a predictor of future success. A team who has gone to more events has a higher chance of success in Atlanta than a team who has gone to fewer. This weighted average, reflects that.

The calculations for these weighted averages are in hidden columns AA-AC on the Master worksheet.

The cool thing about this sheet, is that if you don’t like weights that I’ve assigned for different awards or stats, you can just change them and recalculate everything with your own values.

Thanks to my teammates Steve Rourke, Ian Mackenzie, Geoff Allan and Derek Bessette for all their help putting this together.

If you have any questions, let me know.

See y’all in Atlanta.

I uploaded an old version. I have a newer version with more consistent formatting if anyone wants it. It’s just a but prettier.

Let me know…

This is a very neat database. Good job on it. Very useful. Now, I wanna meet all you guys at the nats :).

Excellent work. It’s great to see a “handicap” on the event. Our team, 842, is ranked 60/72 in the Curie division (or we are in the 16th percentile).
Just the incentive we need:)

Thanks for a well done piece of work!

One thing I forgot to mention, when designing the metric, I wanted to factor in performance in the elimination rounds and draft position. Unfortunately, the data was not available for the first two weekends of the regional season. As a result, if I had included the data, the metric would be skewed towards teams who competed in weeks 3-5.

I’m still kind of confused by the metric. Team 254 has 105+ in Sacramento, but just over 80 in San Jose, even though by the statistics we improved between the regionals (higher mean, lower SD, higher rel ave., higher factor). Is being 2nd seed (in SJ) instead of first (in Sac) or not winning a robot award what causes that?

It was the lack of a technical award which hurt the Poofs in San Jose. From my experiences, I’ve found that more often than not, the technical awards are great predictors of success. I’ve always very happy with the decisions made by the judges. This is why they factor so highly in my metric.

I was shocked when I discovered that the Poofs didn’t win an award in San Jose. It almost made me reconsider the equation. (I figured the Poofs would end up at number 1, just because they’re the best robot I’ve seen.) After further thought, I just figured this lack of an award was an annomaly.

Does anyone have any insight as to why the Poofs didn’t win an award?

If you’re interested, lower the values for the awards on the master metric sheet, and see what happens. I played with various values, and the median I came up with seemed the most reasonable. I’d love to hear some suggestions.

Explain this to me:

My team was the third team in our alliance and we won our regional. How come our partners get a bigger bonus than us. We helped win the regional too and I think we deserve an equal share. We should at least get a bigger bonus that the finalists, or any of the other awards. That’s just not fair.

I’ve noticed that sometimes, judges don’t want to give the same awards to teams year after year. Since the Cheesy Poofs have won just about every award imaginable in the past, this may have played a role.

just pure speculation

We did win Chairman’s Award in San Jose, and it makes sense that the judges would be uneasy about piling awards on any team. Chairman’s is enough award for us any day =).

Could you explain where to change the coefficients in your spreadsheet? I’m awful with Excel, can’t find those hidden numbers.

Found it.

As myself and some members of 115 were chatting Saturday night outside at SVR, Steve Wozniak (A judge at the event) rode by on his Segway and stopped to talk to us. I’m not sure how we got on the subject, but he was telling us that the judges tried to spread awards as much as possible, since there would often be multiple teams that were just as deserving of an award.

He said that a lot of the judges had been at Sacramento, and had seen how a few teams won a bunch of awards, and wanted to avoid doing the same thing at SVR.

Not the topic of the thread, but personally I like this philosophy.

Cory

Alex is right. The judges did not want to give us another award besides chairman’s. We had thought during the awards ceremony that we had a shot at Leadership in Control or Industrial Design, but we got Chairman’s instead (YAY!!! ). Technical Awards are interesting in that a team cannot win two from the same regional in one year. I think Chairman’s got put in this category (maybe I’m wrong…). Are there any chairman’s and technical award winners from the same regional? Also, the judges may have been different in San Jose.

Again, thanks for the compliments.

When creating any sort of statistic, the goal is to be able to make mass generalizations, and be correct most of the time. Stats will never be able to tell the whole story. In some regionals the third robot on a championship alliance is absolutely crucial (See team 229 at the Long Island Regional). In many other cases the role of the third robot is diminished. In my experiences, I’ve found the third robot not to be as strong as the first or second finalist robot. Hence, the statistic was created reflecting this.

No statistic is perfect. I’m sure there will be a few teams at Nationals who do poorly in my rankings who will do well at the event.

The reason I like this statistic, is it rewards robots who put up good numbers (High Averages) and it rewards robots that might not be strong numerically but people still recognize them as good robots (Awards).

Technical Awards are interesting in that a team cannot win two from the same regional in one year.
This is not actually true. If you scan through my spreadsheet, you’ll see a few ocassions of teams winning two technical awards at one event. Just taking a quick glance, Team 456 won both the Delphi and GM award in Texas. I know there are more occurrences than this.

The spreading out of awards at an event usually depends on the judges at the event.

True, the third robot is not always as strong as the first two. BUT, that is already reflected in the qualification matches. Our point total after the qualification matches was lower than our partners. There is no reason why we should be punished again for something unrelated. We are being punished in our winning the regional for our average performance in the qualification matches. One had nothing to do with the other and that should be reflected in the point system.

Whoa now, like Karthik said already, if a team doesnt like how the awards or accolades factor in, they can always change the values of the statistics. This spreadsheet isnt punishing anyone, its only meant as a tool to help. As such, no one is obligated to use it, or believe in its data management system. 188 is ranked pretty low in Archimedes, but that cant be helped. I realise the situation is a little bit different with your robot actually winning a regional and all, but lets try to keep in mind that this is a tool and not an official ranking of the robots coming to the event. Im sure everyone will have an ace or two up their sleeve

Remember what rourke said about the SCAM ratings: the name was intentionally chosen to make sure few people take it seriously. These statistics are the same way: helpful in some ways for determining, roughly, what teams did well at regionals, but not a value judgment of your robot or your team. You aren’t being punished for anything, and teams aren’t being rewarded for anything; it’s just a number. Go in and mess with the spreadsheet on the regional specific pages to change the scores, and adjust it until you feel it’s right, then use it for your scouting purposes.

Just a quick note, Team 967 won and Team 16 have the awards swapped on the spreadsheet (impressive work, btw). We, Team 967, were awarded teh GM ID and BBS teh Driving Tomorrow’s Tech. Not that is it matters much, but just a note of clarity. Sadly, out alliance partners are not in out division…

Ummm, OK I was looking over it again, and…are you sure you got the standard deviations right? There are several SDs that are greater than the mean score, which seems to ring wrong with me (I haven’t taken stats but ± 1 SD is about 2/3 of the data, right?). Did you forget to square root on some of them?

Still very impressive.

The standard deviation is only 2/3 of the data when the data is gaussian (also called normal or bell shaped curve). The score’s of most team’s matches is no where near a bell shaped curve, and so that doesn’t hold. I checked the numbers for team 970 at OH, which was one of the teams where the standard deviation was higher then the mean, and the calculations were correct.

(standard deviation)^2 = (sum(xn - mean)^2)/(n-1)

You can use this webpage to do the calculations for you: Descriptive Statistics: How many items? and this page has explanations of a bunch of different statistical measurments: Descriptive Statistics