Days rest: V2

[kyleskom]

@nkgilley Here is the updated PR with the fixed UO. If you can train the NN UO model and add it here that would be great. I can't see to get tensorflow to train on my Mac currently.

[kyleskom]

This was the fix

[kyleskom]

Fix

[nkgilley]

This is awesome! Glad you were able to get this working. I just retrained the NN model and opened PR #142

[kyleskom]

Gonna give it a few days to see if this looks better. Hard to tell if something still wrong or just actually the predictions for today.

[kyleskom]

NN model still broken. Ill fix that soon.

[kyleskom]

@nkgilley When you trained this new model for the NN did you pull the latest changes from here before training?

[nkgilley]

Yea I'm pretty sure that I did. What looks wrong?

[kyleskom]

It just still seems to be returning overs for all games. Ill have to take another deeper look.

[nkgilley]

from today's games, it's not all overs:

------------Neural Network Model Predictions-----------
Miami Heat vs Memphis Grizzlies (55.2%): OVER 220.5 (52.0%)
Cleveland Cavaliers (55.7%) vs Philadelphia 76ers: UNDER 223 (50.3%)
Chicago Bulls vs Sacramento Kings (57.8%): OVER 236 (49.7%)
Minnesota Timberwolves vs Boston Celtics (61.2%): OVER 233.5 (50.6%)
Houston Rockets vs Los Angeles Lakers (62.8%): OVER 231 (52.8%)
San Antonio Spurs vs Dallas Mavericks (69.8%): OVER 227 (51.7%)
LA Clippers (54.0%) vs Golden State Warriors: OVER 237 (51.6%)

[kyleskom]

So 2 things are happening
1.) The model and prediction is right but is just so close to 50 or it likes the over slightly more
2.) Its still broken

I think with how close he predictions are to 50% im skeptical.

[dimmos1]

can you upload predictions for todays games as mine seems to be playing up

[Marcfeitosa]

Hi Kyle, I've got a quick question. How can i train the model including the 2022 - 2023 season? Cheers Em sexta-feira, 17 de março de 2023 às 09:17:30 BRT, dimmos1 ***@***.***> escreveu: can you upload predictions for todays games as mine seems to be playing up — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[kyleskom]

@Marcfeitosa @dimmos1 Please use the discussions tab. This is for the open PR.

[kyleskom]

Just got back from more travel and going to try to close this this week. Still need to look at the NN model. Also any reason why we don't use the rest days for the ML predictions as well?

[kyleskom]

Im having a hard time getting models I trined to run. What version of python, tensorflow-macos and tensorflow-metal are you using?

[nkgilley]

Python 3.9.6, tensorflow-macos 2.11.0. I'm not using tensorflow-metal it doesn't look like, which probably explains why this computer isn't any faster at this than my 10 year old GPU in my old linux box. I'll look into tensorflow-metal tonight. I know I did before but I think I ran into issues.

[kyleskom]

Im able to train models but when I run them I get an error.

[nkgilley]

I can run the models with and without tensorflow-metal. I can't train with it though. M2 mac mini here.

[nkgilley]

I just got training working with tensorflow-metal==0.5.0 and tensorflow-macos==2.9 suggested by this comment: https://developer.apple.com/forums/thread/721619

I dropped back to the following versions: tensorflow-macos==2.9 and tensorflow-metal==0.5.0. Was using the tensorflow-macos==2.11 and tensorflow-metal==0.7.0 version and just couldn't get things to work. After dropping back I was able to use the GPU and all my validations worked.

I'm able to train and run at much faster speeds now 🚀

[kyleskom]

I was able to fix it. It was due to the ML model being out of date. I think the UO on the NN is also fixed

[kyleskom]

Wanted to loop back to this quickly
#141 (comment)

[kyleskom]

Nvm I think its all good now

[kyleskom]

Looks like days rest is added and tested for both NN and XGB as well as both ML and UO

[kyleskom]

@nkgilley If your still interested in contributing shoot me an email kyleskom@buffalo.edu. Id like to discuss some more things moving forward if your interested.