[Question] How does match simulation work ?

[akasolace]

Hello,

It is a question and not really a bug report but I was not sure where to ask. What methodology are you using to simulate match results ? Is it based on similar match rating on database or are you using a specific formula ?

Thank you

[Blackmorse]

@akasolace Hello, Thanks for question.

I have rather large database of played matches. Currently it is more than 20 millions and it's keep growing every week. So yes, my "simulator" is based on searching of similar matches across all the database.

Sure, this approach have pros and cons, and if anyone have ideas how to improve this I will be glad to consider any suggestions

[Blackmorse]

If you really interested in details, here you can find out how does "similar matches" are defined: https://github.com/Blackmorse/hat-all-stats/blob/master/web/app/databases/requests/matchdetails/SimilarMatchesRequest.scala Will be glad if you will help to improve me this functionality.

Also, feel free to ask questions here on github, or on Hattid telegram channel (you can find it at the top line of every hattid page): https://t.me/hattid_ht

[akasolace]

Thank you for the info .... I was thinking maybe using your huge database of played games to train an AI model to predict game results based on ratings .... what do you think ?

[akasolace]

@Blackmorse actually just discovered something similar has been developed already https://github.com/Ludo2293/ht-ml-predictor. We can ask if nickarana and/or Ludo2293 would be interested in re-running their model on your huge database. Otherwise I might give it a try

[Blackmorse]

@akasolace , thanks for suggestion. Sure it is great idea to use ML for match prediction. It will be great to cooperate with others developers and integrate this solution into the Hattid.

[akasolace]

@Blackmorse ok cool, I will give it a go then. I will come back to you