×
Rock Chalk Talk: Basketball
Anything pertaining to basketball: college, pro, HS, recruiting, TV coverage
Anything pertaining to basketball: college, pro, HS, recruiting, TV coverage
Updated DPPI after tonight's games
- CorpusJayhawk
- Topic Author
- Offline
- Platinum Member
Less
More
- Posts: 1846
- Thank you received: 3634
4 years 4 months ago #23999
by CorpusJayhawk
Don't worry about the mules, just load the wagon!!
KU moves to 3rd overall as 3 undefeated teams go down. Louisville falls from 3rd to 8th. Dayton up to 5th. I have migrated my DPPI over to my website so it is a little easier to read. Here is the link -->>
DPPI
Don't worry about the mules, just load the wagon!!
The following user(s) said Thank You: HawkErrant, Bayhawk, murphyslaw
Please Log in or Create an account to join the conversation.
Share this page:
- Bayhawk
- Offline
- Platinum Member
Less
More
- Posts: 1963
- Thank you received: 850
4 years 4 months ago #24000
by Bayhawk
The end is nothing; the road is all.
-- Jules Michelet
How come KU's mental toughness is zero? :-/
RC
RC
The end is nothing; the road is all.
-- Jules Michelet
Please Log in or Create an account to join the conversation.
- CorpusJayhawk
- Topic Author
- Offline
- Platinum Member
Less
More
- Posts: 1846
- Thank you received: 3634
4 years 4 months ago #24006
by CorpusJayhawk
Don't worry about the mules, just load the wagon!!
Excellent question but not an easy answer. I will try to explain without getting too gory into the guts of it.
The only input into my algorithm is final score (thus scoring margin). So you have to come up with ways of meaningful addressing real world impactful occurrences with only this data. Occurrences such as home, away, neutral for instance. So for starters, you simply take the scoring margin of each team and call that their rating. So ig a team has an average scoring margin of 10, their rating is 10. Simple right? But clearly the flaw is that what if KU plays Duke, Louisville, Maryland and Ohio St. and has an average scoring margin of 2 and LSU, say, plays Houston Baptist, Howard, Delaware St. and Central Connecticut and has an average scoring margin of 4? LSU is rated twice as high as Kansas even though they have played radically divergent strength's of schedule. So you adjust for that by adding in the average scoring margin of each teams opponents. But you don't stop there. You add in the average scoring margin of the opponents of the opponents, the opponents of the opponents of the opponents and so forth. You keep making this adjustment until the adjustment diminishes to insignificance. You also make some sort of meaningful adjustment for home vs, away. Now you end up with a much better "raw predictor" rating. This may result in KU having a rating of 15 and LSU having a rating of -5 for instance, since you have accounted for the quality of opponents. This would be the simplest form of a predictive algorithm and many of them are something very similar to this.
No you can evaluate all of the past games and see how accurate it is. You can also look for non-linearities in the predictive results for things such as games that are played outside of a given standard deviation or games where teams play radically stronger or weaker opponents. So for instance, While the raw predictor says KU has a raw rating of 20 and Milwaukee has a raw rating -15, the projected scoring margin would be 35 points. But what you find is that the relationship to raw rating does not hold linearity as the difference is raw ratings increases. In the real worl this seems explainable because a team will generally not play as hard for as long with the "A": team when they are blowing another team out. So there needs to be a deviation adjustment from linearity. I call this deviation adjustment the "Mental Toughness". It really isn't correlative to the actual mental toughness of the players of the team. It is a mathematical function. As the season progresses, this non-linearity function compresses. If the teams could play 10,000 games it would become a very small number for virtually every team. But with such few data points it can be quite stark. I include it in my output because I use it a lot for testing debugging and experimenting with new routines. I have modified my program a bit this year and whereas the Standard Deviation the last several years has been in the 9.5 range, this season it is 6.2 thus far. I am excited about that but it may change as the season goes along. I am not sure it is due to the improvement of my algorithm or just the games thus far.
Sorry for the gory detail. But you asked.
The only input into my algorithm is final score (thus scoring margin). So you have to come up with ways of meaningful addressing real world impactful occurrences with only this data. Occurrences such as home, away, neutral for instance. So for starters, you simply take the scoring margin of each team and call that their rating. So ig a team has an average scoring margin of 10, their rating is 10. Simple right? But clearly the flaw is that what if KU plays Duke, Louisville, Maryland and Ohio St. and has an average scoring margin of 2 and LSU, say, plays Houston Baptist, Howard, Delaware St. and Central Connecticut and has an average scoring margin of 4? LSU is rated twice as high as Kansas even though they have played radically divergent strength's of schedule. So you adjust for that by adding in the average scoring margin of each teams opponents. But you don't stop there. You add in the average scoring margin of the opponents of the opponents, the opponents of the opponents of the opponents and so forth. You keep making this adjustment until the adjustment diminishes to insignificance. You also make some sort of meaningful adjustment for home vs, away. Now you end up with a much better "raw predictor" rating. This may result in KU having a rating of 15 and LSU having a rating of -5 for instance, since you have accounted for the quality of opponents. This would be the simplest form of a predictive algorithm and many of them are something very similar to this.
No you can evaluate all of the past games and see how accurate it is. You can also look for non-linearities in the predictive results for things such as games that are played outside of a given standard deviation or games where teams play radically stronger or weaker opponents. So for instance, While the raw predictor says KU has a raw rating of 20 and Milwaukee has a raw rating -15, the projected scoring margin would be 35 points. But what you find is that the relationship to raw rating does not hold linearity as the difference is raw ratings increases. In the real worl this seems explainable because a team will generally not play as hard for as long with the "A": team when they are blowing another team out. So there needs to be a deviation adjustment from linearity. I call this deviation adjustment the "Mental Toughness". It really isn't correlative to the actual mental toughness of the players of the team. It is a mathematical function. As the season progresses, this non-linearity function compresses. If the teams could play 10,000 games it would become a very small number for virtually every team. But with such few data points it can be quite stark. I include it in my output because I use it a lot for testing debugging and experimenting with new routines. I have modified my program a bit this year and whereas the Standard Deviation the last several years has been in the 9.5 range, this season it is 6.2 thus far. I am excited about that but it may change as the season goes along. I am not sure it is due to the improvement of my algorithm or just the games thus far.
Sorry for the gory detail. But you asked.
Don't worry about the mules, just load the wagon!!
Please Log in or Create an account to join the conversation.
- JRhawk
- Offline
- Platinum Member
Less
More
- Posts: 704
- Thank you received: 358
4 years 4 months ago #24007
by JRhawk
CJ - Was hoping that someone else the ?, but either I'm the only one who can't get into the DPPI link above, or others are being lazy too.
Anyway, when I click on it (either logged into this board or not, I get a Windows Security Iexplore page saying, "The server kurcc is asking for your user name and password", among other things. I tried my board user name and pw, but doesn't work. Help.
Anyway, when I click on it (either logged into this board or not, I get a Windows Security Iexplore page saying, "The server kurcc is asking for your user name and password", among other things. I tried my board user name and pw, but doesn't work. Help.
Please Log in or Create an account to join the conversation.
- CorpusJayhawk
- Topic Author
- Offline
- Platinum Member
Less
More
- Posts: 1846
- Thank you received: 3634
4 years 4 months ago #24008
by CorpusJayhawk
Don't worry about the mules, just load the wagon!!
JR, Sorry about that. I copied the password link. Many people on this board have a password to my website because of the Rock Chalk Challenge. Here is a link that does not require a password.
DPPI
DPPI
Don't worry about the mules, just load the wagon!!
The following user(s) said Thank You: JRhawk, hairyhawk
Please Log in or Create an account to join the conversation.
- JRhawk
- Offline
- Platinum Member
Less
More
- Posts: 704
- Thank you received: 358
4 years 4 months ago #24009
by JRhawk
Thanks - and yes, a lot easier to read.
Please Log in or Create an account to join the conversation.