Baylus / 2048I

Teaching ML to play 2048 better
MIT License
0 stars 0 forks source link

Resume most recent checkpoints #16

Closed Baylus closed 1 month ago

Baylus commented 1 month ago

Looks good, everything is working perfectly. The only thing I am considering is whether we should even have the weights for the fallback incase we do not have the latest model, which removing the feature would mean we have no need to save off the weights as well as the general model.

Currently the complexity isn't so large in the application that I don't think it is critical to consider removing the feature, and since it only happens at the start, and the saving of the weights happens once every like, 9-12 hours, I don't think that the feature is too consequential for performance.