Fab Mello Ineligible

[Scott Turner]

Are we permitted to make ad-hoc adjustments to our predictions to accomodate this news?  I can't imagine anyone has a model prepared for this situation, but if so I bow to your superiority! :-)

-- Scott

[Lee-Ming Zen]

If your algorithm is capable of ingesting this news, then yes. For example, if your system models the minutes a player plays in a game and there are some overrides to allow you to express "this player will play 0 minutes in his next game" that would be allowed. I would go so far as to say that if you are doing player-level modeling and you have some way to remove his contribution from the team, that would be fine, too. Anything that is human judgment-like would not be okay. For example, if your system predicts margin of victory and you predict Syracuse to win by 5 and you decide that Melo is worth at least 5 points and you change the output. That would not be allowed.

[Scott Turner]

On Tue, Mar 13, 2012 at 4:08 PM, Lee-Ming Zen <lz...@cs.stanford.edu> wrote:

If your algorithm is capable of ingesting this news, then yes. For example, if your system models the minutes a player plays in a game and there are some overrides to allow you to express "this player will play 0 minutes in his next game" that would be allowed. I would go so far as to say that if you are doing player-level modeling and you have some way to remove his contribution from the team, that would be fine, too. Anything that is human judgment-like would not be okay. For example, if your system predicts margin of victory and you predict Syracuse to win by 5 and you decide that Melo is worth at least 5 points and you change the output. That would not be allowed.

Fair enough.  I could add a "Number of missing starters" input to my model and hack in a correction, but I'll restrain myself. 

Has anyone done a player-level simulation for the contest?  I don't recall one, but perhaps I'm mis-remembering.

I've toyed with the idea of building a system to scrape college basketball news, measure the sentiment and subject of each article and then used that to modify/drive predictions.  If I had any assurance that would work, I'd probably give it a shot.  It would be a fun project.

-- Scott