[FRC Blog] Changes to the Standard District Point System

[Hallry]

Posted on the FRC Blog, 1/3/15: http://www.usfirst.org/roboticsprogr...ct-points-2015

Changes to the Standard District Point System

Blog Date: Saturday, January 3, 2015 - 16:24

With Recycle Rush now revealed, some District folks may be wondering how the Standard District Point System will change. Clearly, it can’t remain as it is, as two elements of the current system, performance during qualification rounds, and performance during playoff (formerly elimination) rounds are based on Win-Loss-Tie, which is not part of this season’s game until the finals.

A team of individuals, including two representatives from each District, is currently working on a revised system. This team started work in mid-December, under Non-Disclosure Agreements. The new system is not yet final, but I can tell you this:

• One of our goals is to make as few changes to the system as possible. Only those elements related directly to Win-Loss-Tie are being adjusted
• Another of our goals is to closely approximate the typical points distribution of the current system
  • For the element related to performance during qualification rounds, this means most teams at events will be receiving a moderate number of points, while fewer teams will receive maximum or minimum points. The points awarded for qualification rounds will approximate a normal distribution, as it has historically.
  • For the element related to performance during playoff rounds, this means teams will be awarded points based on advancement to the next round, rather than winning or losing individual matches, with the exception of the final rounds.
• Points assignment for qualification round performance will be based on team rank at the end of qualification rounds, rather than some direct measure of match scores. To earn more points in the system, teams will want to focus on ranking well during events

I am hoping we’ll be able to finalize the system in a few weeks. We’re very close. It will be published as an official part of the FRC Administrative Manual, so all will have easy access to it.

FIRST Championship Waitlist Reopening

As I mentioned during the Founder’s Reception broadcast, we will be hosting 600(!) FRC teams at the FIRST Championship this year, a record number. With that many teams, we will have many waitlist slots available. While we already have a few hundred teams on the waitlist, we want to make sure as many of our veteran teams as possible have a chance to participate in the biggest FRC event ever. So, we’re reopening the waitlist starting at Noon Eastern Time on Monday, January 5th. Also, in a change to what I said about the waitlist in an earlier blog here, we have decided to allow teams that attended the FIRST Championship last year, 2014, to add their names to the list!

Teams will receive a number of ‘entries’ to get selected from the waitlist equal to 2015 minus the last year they attended Championship (or their rookie year, or when they most recently returned as an active team, whichever year is most recent). As examples, a team that attended last year will get one entry in the waitlist drawing, a team that last attended in 2013 will get two, and a team that has never attended but has been continuously active since 2005 will get ten.

Rookies will still not be eligible for the waitlist, but we’re looking forward to a number of Rookies earning their way to Championship based on their performance at events.

Also, the waitlist won’t be closing until April 6th, so you’ll have plenty of time to throw your hat in the ring.*

Frank

*TIL http://www.phrases.org.uk/meanings/t...-the-ring.html

[Nathan Streeter]

Guessing no one has any news regarding the changes to the Standard District Point System based on this game's unique structure?

[Basel A]

Originally Posted by Nathan Streeter

Guessing no one has any news regarding the changes to the Standard District Point System based on this game's unique structure?

I heard a rumour the replacement for WLT will be some sort of linear points structure based on final rank. As far as I know, no information has been officially released.

[Christopher149]

Originally Posted by Nathan Streeter

Guessing no one has any news regarding the changes to the Standard District Point System based on this game's unique structure?

According to Q222, they've figured out the structure, and I expect it will be released by Wednesday of this week if they keep to 2 weeks.

[Navid Shafa]

Originally Posted by Basel A

As far as I know, no information has been officially released.

Also not offical, but I remember this vaguely:

It's a bell curve, with a normalized amount of points at the upper limit to keep the values within a similar range as last year's potential on field point structure.

[piersklein]

I believe Frank talks about it in the GameSense Holiday Special http://https://www.youtube.com/watch?v=OFQPemOh4bI

[AGPapa]

Originally Posted by Navid Shafa

Also not offical, but I remember this vaguely:

It's a bell curve, with a normalized amount of points at the upper limit to keep the values within a similar range as last year's potential on field point structure.

I checked out the past distribution of WLT points to see if it actually was normal in the past. I've attached a graph of the distribution of WLT points in MAR for the past three seasons. (And also a comparison to a normal distribution with the parameters observed the past three years).

Interestingly, no MAR team in the past few years went 12-0-0 at a district event. The closest teams were 341 at Hatboro 2012 and 2590 at Hatboro 2013 (both went 11-0-1). Using a normal distribution there should only have been ~1.3 teams over the past few years to get the full 24 WLT points. The new system will probably give the top seed 24 points though, changing the system slightly from either a normal distribution or the actual distribution.

Additionally the normal distribution would give more teams 0 WLT points than has happened the past few years. Two teams have completed an event with 0 WLT points over this timespan and the normal distribution says that ~1.5 teams should have received 0 points during those seasons. I wonder if this new system will give points to a team who seeded last.


On a different note, I'm mildly upset that FIRST hadn't thought about district events when they wrote the past two manuals. This WLT issue should have been addressed and figured out before kickoff. In this Frank Answers Friday post Frank said that the rules governing advancement to the District Championship and the World Championship would be in the 2015 manual. They are not.

[Caleb Sykes]

Originally Posted by AGPapa

I checked out the past distribution of WLT points to see if it actually was normal in the past. I've attached a graph of the distribution of WLT points in MAR for the past three seasons. (And also a comparison to a normal distribution with the parameters observed the past three years).

What did you use as your standard deviation for the normal distribution? Did you use the standard deviation of the observed data from previous years?

[Basel A]

Originally Posted by Navid Shafa

Also not offical, but I remember this vaguely:

It's a bell curve, with a normalized amount of points at the upper limit to keep the values within a similar range as last year's potential on field point structure.

If I were to pick, I'd probably pick something like that. I was a bit put off when I was told it might be linear. We'll see, I guess.

[AGPapa]

Originally Posted by Caleb Sykes

What did you use as your standard deviation for the normal distribution? Did you use the standard deviation of the observed data from previous years?

Yes, the average of the previous data was 11.9 and the standard deviation was 4.44.

[Caleb Sykes]

Originally Posted by AGPapa

Yes, the average of the previous data was 11.9 and the standard deviation was 4.44.

Interesting. The first graph you have there seems to have the normal distribution shifted to the right of the observed data. My guess is this is due to how you put the normal distribution into bins for the histogram.

It looks like you used bins with ranges of [2n, 2n+2) where n is the nth bin from the right. For example, for the 10/11 bin, did you take the normal distribution from [10,12) for the bin?

If you want to get the normal curve data to better align with the observed data on this histogram, try using bin ranges of [2n-1, 2n+1). That way, each bar is centered on it's respective W-L record. This doesn't account well for ties, but there have been few enough ties in the last few years that this effect should be negligible.

[AGPapa]

Originally Posted by Caleb Sykes

It looks like you used bins with ranges of [2n, 2n+2) where n is the nth bin from the right. For example, for the 10/11 bin, did you take the normal distribution from [10,12) for the bin?

I believe that it looks "shifted" due to ties and odd numbers of points. In reality very few teams get ties, but the normal distribution doesn't know this. I put 10 and 11 in the same bin so that there wouldn't be dips at every point. The actual distribution is also shifted down, (there are more really bad teams than really good teams). Again, the graph is just to get an idea of how the rule changes will affect things.

For the 10/11 bin I added the pdf of 10 with the pdf of 11; I'm not sure how this relates to a continuous function. I've attached the spreadsheet I used for you to look at. It's a little messy as I did not originally intend to upload it, however I added some labels and it should be clear enough.

[Caleb Sykes]

Originally Posted by AGPapa

I believe that it looks "shifted" due to ties and odd numbers of points. In reality very few teams get ties, but the normal distribution doesn't know this. I put 10 and 11 in the same bin so that there wouldn't be dips at every point. The actual distribution is also shifted down, (there are more really bad teams than really good teams). Again, the graph is just to get an idea of how the rule changes will affect things.

For the 10/11 bin I added the pdf of 10 with the pdf of 11; I'm not sure how this relates to a continuous function. I've attached the spreadsheet I used for you to look at. It's a little messy as I did not originally intend to upload it, however I added some labels and it should be clear enough.

It was clear enough.

Here is the same histogram using the method I previously described. It has less than half of the variance of the other method, so it is likely a better fit. I also used a continuous integration of the normal distribution instead of using discrete points as an approximation. So that may also have had a small effect.

[Caleb Sykes]

Here is the excel spreadsheet I used in case anyone is interested.