I evaluated the performance of the model by backtesting it on the 2022-23 season. (so far)

[drdml]

I backtested the "XGBoost_74.9%_ML-2.json" model on the 2022-23 season.

I downloaded data from teams using the Get_Data module, and then created a new function to acquire all historical odds from FanDuel using the SbrOddsProvider library. Next, I executed the Create_Games module to create the 2022-23 dataset.

The 2022-23 dataset now contains two additional columns: home_team_odds and away_team_odds. These are necessary for the expected value calculation.

Once all steps were completed, I ran the Test_Current_Season script. Rather than returning only the winner prediction, I made a few modifications to the code so that it now returns the expected value, the probability, and the winner prediction. It only returns the team with the highest expected value.

For a $1 dollar per bet, the graph starts one month after the season start.

I utilized WPS Spreadsheets, and encountered a few difficulties while attempting to upload to Google Sheets and online Excel. I will provide the link to the .xlsx file, but here is a picture that displays all the information on the spreadsheet:

Link to .xlsx file: https://mega.nz/file/YNgzxBwa#fo3Sp10MeefLl22z6qNAU5e2s7I4Q9MaOPd9s2ziAn8

[kyleskom]

Thanks for sharing this. I think seeing stats like this a helpful and allow us to visualize different strategies based on EV value or win probabilities and make bets accord to certain thresholds. One question, for the second column prediction they are all 1, is that expected?

[drdml]

Yes, I did not want to leave it blank, so I filled it with a one to see what the result would be if I had always bet on the home team.

[kyleskom]

Thats concerning, will take a look at it deeper and maybe start training some new models.

[trevon55]

This is an interesting project, and perhaps I have missed something, but appears you are not using player level data or models to make these predictions? I also cannot see rest, travel, consecutive home and road games, and some other important metadata being accounted for.

Edit: Reading further I can see that such queries have been made previously by others, and that time (as always) is a constraint. This is a great piece of work considering it started more as a learning exercise than anything else. However, I wouldn't recommend anybody bet on predictions made without accounting for a lot more then the moneyline model you have currently does, there is essentially no way you can have positive EV with the dataset currently in use.

[kyleskom]

I would agree. Treat this more as 1 tool in a tool belt as opposed to a single point of truth. Do hope to bring in more data you mentioned and player data down the road!

[JackPayne123]

I would love to see this done for O/U predictions maybe something could work with using the confidence percentage?

[drdml]

https://docs.google.com/spreadsheets/d/1LxPq1ZF3k6BreB8p50vESP-7BTZXVrFs6abU7qxaq2c/edit#gid=1616360111

found this on Reddit, it's not mine. someone shares the picks of the model every day,

[nealmick]

I talked to that guy, seems cool, should probably credit project in posts.

[kyleskom]

Interesting stats thanks for sharing. Ya I saw his Reddit posts as well. Some credit or just at least directing to my repo would be nice but all good.

[obtim]

https://docs.google.com/spreadsheets/d/1LxPq1ZF3k6BreB8p50vESP-7BTZXVrFs6abU7qxaq2c/edit#gid=1616360111

found this on Reddit, it's not mine. someone shares the picks of the model every day,
Please, Share the link to his posts on reddit. Can't find it :(

[FutureTech22]

is there anything like this for the neural network model? or just the XG-boost model?

[drewhb]

@notfermat How/where did you get 74.9% model? I don't see any models with that percentage and after training I don't get any with that high a percentage. Maybe its just a matter of when you trained the data for the season since it was a few months ago?

[dunaevv]

@notfermat How/where did you get 74.9% model? I don't see any models with that percentage and after training I don't get any with that high a percentage. Maybe its just a matter of when you trained the data for the season since it was a few months ago?

I have 69.1% ML. Just keep training bro!

[rpassaro]

By this do you mean we just keep running the training commands? How do you know the %?

[AaronAshery]

I backtested the "XGBoost_74.9%_ML-2.json" model on the 2022-23 season.

I downloaded data from teams using the Get_Data module, and then created a new function to acquire all historical odds from FanDuel using the SbrOddsProvider library. Next, I executed the Create_Games module to create the 2022-23 dataset.

Is this in the repo? if not would you mind sharing this function? I'd like to maybe test out how it would do betting based on EV and kelly

[MMSahin]

@lpUUzVRWtKgp can you please share the function in the repo or somehow else ? :)

it would be great if you could do this in comprahension between NN and XGB Model.

Big thanks!

[Hasan-Reza]

Can someone please explain how to get the Test_Current_season.py file running. It is trying to read from a dataset that does not exists in the repo nor is it being created.

[dunaevv]

I backtested the "XGBoost_74.9%_ML-2.json" model on the 2022-23 season.

I downloaded data from teams using the Get_Data module, and then created a new function to acquire all historical odds from FanDuel using the SbrOddsProvider library. Next, I executed the Create_Games module to create the 2022-23 dataset.

The 2022-23 dataset now contains two additional columns: home_team_odds and away_team_odds. These are necessary for the expected value calculation.

Once all steps were completed, I ran the Test_Current_Season script. Rather than returning only the winner prediction, I made a few modifications to the code so that it now returns the expected value, the probability, and the winner prediction. It only returns the team with the highest expected value.

For a $1 dollar per bet, the graph starts one month after the season start.

I utilized WPS Spreadsheets, and encountered a few difficulties while attempting to upload to Google Sheets and online Excel. I will provide the link to the .xlsx file, but here is a picture that displays all the information on the spreadsheet:

Link to .xlsx file: https://mega.nz/file/YNgzxBwa#fo3Sp10MeefLl22z6qNAU5e2s7I4Q9MaOPd9s2ziAn8

Ty, now I have 74.8% for XGBoost Model ML and 62.2 for OU.

[mohcly]

the link is dead bro, please can I have a new one?

[rpassaro]

Could you share the 62.2% file please?