You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In order to measure the objective performance of kanachan, a baseline mahjong AI model that satisfies the following conditions will be needed:
it performs to a certain degree,
the model is publicly available, and
its performance on major online mahjong platforms such as Tenhou and Majsoul is roughly known.
I am currently planning to use akochan for performance measurement as a baseline satisfying the above conditions. However, since akochan is based on the mjai protocol but kanachan's mjai support has not been implemented yet, the objective performance of kanachan is still unknown. Please wait a little longer for reports on kanachan's objective performance.
Once I implement performance comparison with akochan, I will publish it in this repository so that you can use it for objective performance measurement.
I couldn't find a place to ask, so I'm asking you here.
The text was updated successfully, but these errors were encountered: