Machine learning on Tennis Bets !

• create a fresh virtual environment : py -m venv env
• on visual studio code you might have to elevate your prompt : Set-ExecutionPolicy Unrestricted -Scope Process
• activate your environment : .\env\Scripts\activate
• active your virtual environment : py -m pip install -r requirements.txt

• on visual studio code you might have to elevate your prompt : Set-ExecutionPolicy Unrestricted -Scope Process
• activate your environment : .\env\Scripts\activate
• go in the streamlit folder : cd .\streamlit_app\
• launch Streamlit : streamlit run app.py

• This is optional and not recommended (very long) as all datas are already scrapped and in folder
• Data is coming from multiples sources :
  • https://www.atptour.com for tournaments, matchs, matchs stats and players
  • http://tennis-data.co.uk/alldata.php for matchs odds and players ranking in time

What generate_training_dataset does :

1. Merge scrapped atp match data + scrapped atp players data + scrapped tennis-data.co.uk alltogeter per match
2. winner player is rename as player 1, loser player is rename as player 2
3. 50% of the data are randomly schuffled (inverting player 1 and player 2) so that winner may be 1 or 2

scrapping :

• get odds

Get players :

• performance per tournament (habits to go in finals)
• progression over last X month / last X tournaments
• timeplay regularity
• overall performance over higher ranking
• overall performance over below ranking
• did this tournament already (yes :1 : no : 0)
• lefty / righty
• intéger la fréquence moyenne de blessures annuelles dans le roi (le ROI peut il absorber cette perte ?)

Get next tournaments:

• create a predict function that only use a date parameter to get the next match on the internet

Code in V1 was initally a fork from : https://github.com/edouardthom/ATPBetting Code in V2 is using scrapping code from : https://github.com/serve-and-volley/atp-world-tour-tennis-data