Page 2 of 2 FirstFirst 1 2
Results 21 to 34 of 34

Thread: The impact of the ULM series ...

  1. #21
    Join Date
    May 2002
    Location
    Austin, Texas
    Posts
    5,020

    Default Re: The impact of the ULM series ...

    Quote Originally Posted by Cajunsmike View Post
    This situation reinforces my belief that the RPI is too heavily schedule based. You only control 25 percent of results from your play. Instead of a 25-50-25 basis it should be 40-35-25. You would still have schedule as a significant component in that system but not nearly to same degree. I bet the major schools would fight that change.
    Certainly winning is under emphasized in the RPI (just over 25%) and SOS is over emphasized (just over 75%) ... one of the principal complaints I have been voicing for years. But modifying the weights will not solve the problems. A core problem is that two levels of interconnectivity is simply not enough to capture real SOS. I have provided many examples over the years supporting why. The problem with RPI is structural and cannot be fixed with tweaks.

    Brian

  2. #22
    Join Date
    Dec 2002
    Location
    Breaux Bridge
    Posts
    3,854

    Default Re: The impact of the ULM series ...

    Quote Originally Posted by GoneGolfin View Post
    Not overall SOS ... this is 33. NC SOS. See above rankings I provided.

    Brian
    You are correct. I stand corrected.

  3. #23
    Join Date
    Feb 2017
    Posts
    349

    Default

    Quote Originally Posted by GoneGolfin View Post
    Certainly winning is under emphasized in the RPI (just over 25%) and SOS is over emphasized (just over 75%) ... one of the principal complaints I have been voicing for years. But modifying the weights will not solve the problems. A core problem is that two levels of interconnectivity is simply not enough to capture real SOS. I have provided many examples over the years supporting why. The problem with RPI is structural and cannot be fixed with tweaks.

    Brian
    That's exactly what I've been saying for a long time.




    What a burger!

  4. #24
    Join Date
    May 2002
    Location
    Austin, Texas
    Posts
    5,020

    Default Re: The impact of the ULM series ...

    Quote Originally Posted by whataburger View Post
    That's exactly what I've been saying for a long time.


    Actually, you are right ... I believe you taught me that. :-)

    Brian

  5. #25
    Join Date
    Feb 2017
    Posts
    349

    Default

    Quote Originally Posted by GoneGolfin View Post
    Actually, you are right ... I believe you taught me that. :-)

    Brian
    Brian, my brain would short-circuit before understanding half of what you do about rpi, among other things.

    The NCAA should just hire you to fix everything. Let me know if you want me to put in a good word for you.


    What a burger.

  6. #26
    Join Date
    Apr 2011
    Location
    Lafayette
    Posts
    118

    Default Re: The impact of the ULM series ...

    Brian. What would be the system you would set instead of the current RPI System?


  7. #27
    Cajunsmike is offline Ragin Cajuns of Louisiana Ragin' Cajuns Greatest Fan Ever
    Join Date
    Jan 2003
    Posts
    5,648

    Default

    Quote Originally Posted by GoneGolfin View Post
    Certainly winning is under emphasized in the RPI (just over 25%) and SOS is over emphasized (just over 75%) ... one of the principal complaints I have been voicing for years. But modifying the weights will not solve the problems. A core problem is that two levels of interconnectivity is simply not enough to capture real SOS. I have provided many examples over the years supporting why. The problem with RPI is structural and cannot be fixed with tweaks.

    Brian
    My suggestion may not totally fix the issue but it would be an improvement over what we have now

  8. #28
    Join Date
    Sep 2002
    Posts
    1,793

    Default Re: The impact of the ULM series ...

    Question... Cajuns play @ GA Southern Wed.... and Cajuns will be the home team. Does the home/road "bonus" for W's and L's come in play here? Or does every win/loss = 1? Thx!


  9. #29
    Join Date
    Feb 2017
    Posts
    349

    Default

    Quote Originally Posted by cajundiehard View Post
    Question... Cajuns play @ GA Southern Wed.... and Cajuns will be the home team. Does the home/road "bonus" for W's and L's come in play here? Or does every win/loss = 1? Thx!
    No bonuses. All games are neutral.

  10. #30
    VObserver's Avatar
    VObserver is online now Ragin Cajuns of Louisiana Ragin' Cajuns Greatest Fan Ever
    Join Date
    Jun 2002
    Location
    Abbeville, LA
    Posts
    5,115

    Default Re: The impact of the ULM series ...

    Quote Originally Posted by IHateLaState View Post
    I found this especially interesting. Despite the bellyaching we have some quality wins this season and yes some mind-numbing losses and a tie (taking a lead into the 9th). 3-3 v Top 25... I guess that is why I keep saying if we are a bubble (and that assumes the 6-1 finish) I think we have a very good shot.. Brian who knows more than I could ever dream of knowing calls it "in consideration"... I maintain we have built a rep over the past few years and I think 6-1 gets us in... We shall see
    Tell that to the softball selection committee. Reputation doesn't mean anything unless you are in a power conference.

  11. #31
    Join Date
    Dec 2012
    Posts
    4,335

    Default Re: The impact of the ULM series ...

    Got this link from the News Star. Pretty brutal, just look at the topics and the common theme.

    http://www.thenewsstar.com/sports/ulm/


  12. #32
    Join Date
    May 2002
    Location
    Austin, Texas
    Posts
    5,020

    Default Re: The impact of the ULM series ...

    Quote Originally Posted by Cajun Troll View Post
    Brian. What would be the system you would set instead of the current RPI System?
    It probably does not exist. But at its core, it would need to be a system that takes into account many levels of interconnectivity. This is instrumental in producing a system that can calculate an accurate SOS (using win/loss results). The current system only accounts for opponents and opponents' opponents. So, as an example ... it is not enough to determine the actual strength of the opponents on your schedule. Team A (your opponent) might play some weak opponents and defeat those opponents ... but those opponents also play weak schedules and they have success against them. The result will be a highly exaggerated OWP and OOWP (and thus SOS) contribution to your RPI ... because it did not go nearly deep enough to determine the truth. The converse is also true ... and just as important. Also, a sound system will not need all of the gimmicks added to it (bonuses, penalties, etc.) ... as well as redundant criteria in the selection processes.

    This type of recursive system I would propose would also remove the arbitrary use of weights to the components currently assigned by humans. Instead, the system would learn these weights or coefficients. I am not sure whether this would be best implemented using a linear model (maybe using logistic regression) or a non-linear model ... but both could be implemented and tested.

    The beauty of this approach is that we have massive amounts of good results data to both train and eval/test the model. So, obviously I believe that utilizing a machine learning approach (really just statistical learning in software) is the way to go. Maybe once I retire from my current AI work, it would be a fun project.



    --

    Now, there are a few systems out there that could be plucked (certainly not as good as the approach I would take above) that would improve significantly over RPI. One of them would be ISR. I would consider this as a decent stopgap solution.

    Brian

  13. #33
    Join Date
    Nov 2010
    Location
    broussard,la
    Posts
    4,474
    Blog Entries
    3

    Default

    Quote Originally Posted by GoneGolfin View Post
    It probably does not exist. But at its core, it would need to be a system that takes into account many levels of interconnectivity. This is instrumental in producing a system that can calculate an accurate SOS (using win/loss results). The current system only accounts for opponents and opponents' opponents. So, as an example ... it is not enough to determine the actual strength of the opponents on your schedule. Team A (your opponent) might play some weak opponents and defeat those opponents ... but those opponents also play weak schedules and they have success against them. The result will be a highly exaggerated OWP and OOWP (and thus SOS) contribution to your RPI ... because it did not go nearly deep enough to determine the truth. The converse is also true ... and just as important. Also, a sound system will not need all of the gimmicks added to it (bonuses, penalties, etc.) ... as well as redundant criteria in the selection processes.

    This type of recursive system I would propose would also remove the arbitrary use of weights to the components currently assigned by humans. Instead, the system would learn these weights or coefficients. I am not sure whether this would be best implemented using a linear model (maybe using logistic regression) or a non-linear model ... but both could be implemented and tested.

    The beauty of this approach is that we have massive amounts of good results data to both train and eval/test the model. So, obviously I believe that utilizing a machine learning approach (really just statistical learning in software) is the way to go. Maybe once I retire from my current AI work, it would be a fun project.



    --

    Now, there are a few systems out there that could be plucked (certainly not as good as the approach I would take above) that would improve significantly over RPI. One of them would be ISR. I would consider this as a decent stopgap solution.

    Brian

    That's a lot of information to digest the morming after 2 6 packs of Ipa 's. Lol

  14. #34
    Join Date
    Apr 2011
    Location
    Lafayette
    Posts
    118

    Default Re: The impact of the ULM series ...

    Thanks Brian.


Page 2 of 2 FirstFirst 1 2

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •