Skip to content

Andrew, Sid - Final project progress - #48

Open
sidlamsal wants to merge 1 commit into
ernbilen:mainfrom
sidlamsal:final
Open

Andrew, Sid - Final project progress#48
sidlamsal wants to merge 1 commit into
ernbilen:mainfrom
sidlamsal:final

Conversation

@sidlamsal

Copy link
Copy Markdown

@ernbilen

Copy link
Copy Markdown
Owner
  • Nice use of screenshots. It reminded me of: https://imgflip.com/memegenerator/120396944/Crazy-Board
  • Make sure you don't use any future predictors in training. I think we did something similar to this in Andrew's independent study. When the season starts, playoff reaching likelihood should be a function of roster quality only, maybe some influence from previous season performance, but as teams play more games, probabilities should get more accurate.
  • You should create a "thrill" metric that is a function of the point spreads. That is, if a team wins the season 10 games in advance, there wouldn't be much thrill then. Once you have this, you can try simulating game results based on say team Elos with different season length, and maximize thrill using optimal season length.

@ernbilen

ernbilen commented May 8, 2025

Copy link
Copy Markdown
Owner
  • Great use of pics and screenshots.
  • Ryan and grace in the other section worked on baseball playoff prediction, so considering chatting and comparing your results with them.
  • You unpacked more why model is not learning more but I was a bit confused. I was thinking your first graph used rolling average, but after seeing your second graph I realized your first graph is something else. In other words, no matter what happens, I think the model should be able to predict more precisely given 2 games for every team in the season vs 20 games. A bit more clean-up here would be useful. As a sanity check, the week where final games are played should have very high precision.
  • You can try and show how elo ratings would compare with your model's accuracy.
  • Great team level suggestions, comparing Celtics and spurs.
  • Nice long run use of team data.
  • Nice simulation too!
  • Can also consider comparing your model to bet odds.
  • Consider expanding your data to player level for more precise suggestions. That is, unpacking beyond things that'd help a coach. Eg, they know they need to win more, but how?
  • Great ethics discussion. Use of quotes from people in charge of running the league was really awesome.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants