search

miroesli / psscscs

Reinforcement Learning Battlesnake
MIT License
2 stars 1 forks source link

API #22

Closed Krocodial closed 4 years ago

Krocodial commented 4 years ago

main will load a model if it exists or create a new one and write to file. The model is saved under data/ by default although we could change that to model/ I just forget what each is for. I considered including training in main to allow training via the API, but we can discuss if we want that or not

miroesli commented 4 years ago

Main comments:

main will load a model if it exists or create a new one and write to file.

I think it should only load a model from models, otherwise choose a random direction, nothing should be created in the main because execution time is a concern. It should be thought of as the file that gets run by the battlesnake servers.

The model is saved under data/ by default although we could change that to model/ I just forget what each is for.

Should be model, data is for logging information during training sessions

I considered including training in main to allow training via the API, but we can discuss if we want that or not

I vote that we definitely don't train in main.

miroesli commented 4 years ago

Previous comments resolved over call with @Krocodial . Loading models from model, and if it doesn't exist create a new one which will select a random direction and optionally train based on input (though this will likely not be effective compared to competing locally due to delay between moves on website)