i want all of the coordinates of the towers, i want to get a deep learning algoritme to do the job for me an every time get better
all of the coordinates of the towers
This can be done by looping through the x and y coordinates in
range(game_state.ARENA_SIZE). Check to see if the location is in the arena bounds and if the game map contains a stationary unit at that location (both member function of the game map, with I recommend getting familiar with). If so, add it to your list.
a deep learning algorithm
This is quite more involved, and if asking the above question, I recommend making a regular, handcrafted algo while you’re still familiarizing yourself with the game. I tried my hand at it here, and I added in the Slack channel later on the following (that would make it easier, IMO), but bear in mind that I haven’t tried this:
If you want to avoid making your own game engine to play hypothetical games, you can use the
run_match.shscript to locally play games using the engine.jar in the starter kit. I didn’t bring this up in the DQN post because I wanted to make my own, but theoretically if you could find some way for the training environment to talk to algo that’s running and use it to train, that should be sufficient. I believe
algo_core.pyalso receives one frame for the end of the game; if that’s so you could very much make an RL environment that uses the existing engine and use that frame to notify the environment that the game has reset.