I am off to NW Regionals this weekend and here is the short explanation of what I am going to be watching. I'm still not exactly sure because the schedule is a bit...hm, how to say this graciously...unfriendly to spectators. Also, Mizu and I are going together and since we have three little kids we don't actually get to talk to each other about what we are doing. We know we are getting in the car after work on Friday and driving back on Sunday, but that's about it. On to the games.

Saturday

1. The two good open games are Revolver-Rhino and Sockeye-Furious (classic!). They are scheduled simultaneously (are you kidding me!?) at 9AM (are you kidding me?!?) Ugh. I am not really sure what I will do about these. I had really hoped to spend Saturday morning in Seattle hitting the old breakfast haunts, but damn, Sockeye-Furious.

2. Zeitgeist-Further. Further has never beaten Zeitgeist and if they want to make Nationals, they would do well to win here. The top two teams in each pool are guaranteed two shots at making the Show, while everyone else must battle it out for the game-to-go. A big chunk of Further played for a team I know pretty well, so I am a bit partial.

3. If I skip the men's games in the morning, I will probably stay and watch the last round on the women's side which will feature Riot-Zeitgeist and Traffic-Underground.

Sunday

1. Round 1: Men's semis. This should feature some combination of Sockeye, Furious, Rhino and Revolver. The chalk has Sockeye over Rhino and Revolver over Furious. Mandatory watching.

2. Round 2: Women's final. Should be Riot-Fury. Adjacent fields sport the 3-4 (winner to Natties) and 5-6 games (loser goes home).

3. Round 3: This is the tough one. It is the men's final. It is also the men's 3-4 game with the loser going home. It is the women's game-to-go. I will watch Sockeye wherever they are and Further if they are in the game-to-go (otherwise they are finished). Fortunately, all these games are scheduled onto the same quadrant of fields.

4. Round 4: Men's game to go. There is no feeling in the world like watching the game-to-go after you've already qualified.

## Friday, September 30, 2011

## Thursday, September 22, 2011

### Why Mini?

So big thanks to everyone who helped with the nerdy challenge. You all saved me a ton of time. In thanks, I'd like to explain the frisbee reasons I was curious.

Mini is an awesome tool for development of individual skills. Unlike 7s, where a young player might touch the disc only 4 or 5 times a practice, Mini affords them a countless touches in a very short period of time. Touches, marking, defending, reading...every essential skill is practiced at game speed in a game like setting at a rate you can't match in a full-sided scrimmage.

I was interested in using it as a tool at the team level. In particular, I wondered if it would be possible to set a possession goal and use mini scores as a way to evaluate this goal. What I mean is, could you record the scores to a whole bunch of mini games and then go back and figure out what the possession rate was? Actually, I knew you could do this work, but I needed the algorithm to figure out the rates. That was the reason for the nerdy challenge.

Using Alpha Chen's probability generator, I cranked out possibilities established expected values and was able to come up with an expected value number for each possession probability. (The spreadsheet is here.) There is a bit of inaccuracy because Alpha Chen's simulator is a Monte Carlo generator and comes up with different values on each run, but from a frisbee standpoint, it doesn't need to be exact.

Here's how it would work: You play mini. Everyone keeps a running total of their score for all their games and how many games they played. (Golden Goal scenarios still 'score' as -1, -2.) When you are done, you add them all up and divide by how many games where played. Compare this number to the chart and voila! You know how you did on possession percentage.

10% = -1.4

20% = -1.2

30% = -1.0

40% = -0.5

50% = 0.1

60% = 0.7

70% = 1.2

80% = 1.4

Last thought. Some of the direction for this thinking came from Anson Dorrance's Vision of a Champion. In particular, Ch 12 and 13.

I think my reasoning is solid, but please, if I messed up, let me know. Thanks!

Mini is an awesome tool for development of individual skills. Unlike 7s, where a young player might touch the disc only 4 or 5 times a practice, Mini affords them a countless touches in a very short period of time. Touches, marking, defending, reading...every essential skill is practiced at game speed in a game like setting at a rate you can't match in a full-sided scrimmage.

I was interested in using it as a tool at the team level. In particular, I wondered if it would be possible to set a possession goal and use mini scores as a way to evaluate this goal. What I mean is, could you record the scores to a whole bunch of mini games and then go back and figure out what the possession rate was? Actually, I knew you could do this work, but I needed the algorithm to figure out the rates. That was the reason for the nerdy challenge.

Using Alpha Chen's probability generator, I cranked out possibilities established expected values and was able to come up with an expected value number for each possession probability. (The spreadsheet is here.) There is a bit of inaccuracy because Alpha Chen's simulator is a Monte Carlo generator and comes up with different values on each run, but from a frisbee standpoint, it doesn't need to be exact.

Here's how it would work: You play mini. Everyone keeps a running total of their score for all their games and how many games they played. (Golden Goal scenarios still 'score' as -1, -2.) When you are done, you add them all up and divide by how many games where played. Compare this number to the chart and voila! You know how you did on possession percentage.

10% = -1.4

20% = -1.2

30% = -1.0

40% = -0.5

50% = 0.1

60% = 0.7

70% = 1.2

80% = 1.4

Last thought. Some of the direction for this thinking came from Anson Dorrance's Vision of a Champion. In particular, Ch 12 and 13.

I think my reasoning is solid, but please, if I messed up, let me know. Thanks!

## Thursday, September 8, 2011

### Skyd + Nerdy Challenge

If you haven't seen already, I have moved over to Skyd. I am going to keep this space open for little odds and ends and things that don't quite fit or are a bit less formal than I intend the Skyd writing to be.

Here is the nerdy challenge: I am looking for a function where the input is the chance of a possession ending in a score (p). Obviously, 1 - p is the chance that a possession ends in a turnover. Then output is the probability of each of the six possible endings to a game of mini. The three 'wins' being: 3~1, 3~0, 3~-1 and the three 'losses': 1~-2, 0~-2 and -1~-2. Please ignore the Golden Goal variation for right now.

I brute forced it for p=0.5. This had the nice property that 1-p=0.5 as well, so the outcome tree was essentially symmetric and the probabilities easier to manage because I just used negative powers of 2. I quit after 5 generations and then estimated the remaining 20% of the probability based on the observed trends. Here is what I got:

3~1: 4.4%

3~0: 23.6%

3~-1: 18.2%

Winning: 46.2%

1~-2: 9.2%

0~-2: 18.6%

-1~-2: 25.6%

Losing: 53.4%

This was a bit surprising. I expected the chance of losing to be much higher. My intuition told me that it was two steps to lose and three steps to win, so it should divide out on a two-thirds, one-third ratio. But it turns out that you always have a chance to win, but you don't always have a chance to lose.

Help? Ideas? There has to be an easier way.

Here is the nerdy challenge: I am looking for a function where the input is the chance of a possession ending in a score (p). Obviously, 1 - p is the chance that a possession ends in a turnover. Then output is the probability of each of the six possible endings to a game of mini. The three 'wins' being: 3~1, 3~0, 3~-1 and the three 'losses': 1~-2, 0~-2 and -1~-2. Please ignore the Golden Goal variation for right now.

I brute forced it for p=0.5. This had the nice property that 1-p=0.5 as well, so the outcome tree was essentially symmetric and the probabilities easier to manage because I just used negative powers of 2. I quit after 5 generations and then estimated the remaining 20% of the probability based on the observed trends. Here is what I got:

3~1: 4.4%

3~0: 23.6%

3~-1: 18.2%

Winning: 46.2%

1~-2: 9.2%

0~-2: 18.6%

-1~-2: 25.6%

Losing: 53.4%

This was a bit surprising. I expected the chance of losing to be much higher. My intuition told me that it was two steps to lose and three steps to win, so it should divide out on a two-thirds, one-third ratio. But it turns out that you always have a chance to win, but you don't always have a chance to lose.

Help? Ideas? There has to be an easier way.

Subscribe to:
Posts (Atom)