import gym_microrts
from gym_microrts.envs.vec_env import MicroRTSGridModeVecEnv
from gym_microrts import microrts_ai
from gym.envs.registration import register
from gym_microrts import Config
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torch.distributions.categorical import Categorical
try:
env = MicroRTSGridModeVecEnv(
num_envs=1,
render_theme=2,
ai2s=[microrts_ai.passiveAI],
map_path="maps/16x16/basesWorkersTestAttack16x16.xml",
reward_weight=np.array([10.0, 1.0, 1.0, 0.2, 1.0, 4.0]))
# env = gym.make('MicrortsDefeatCoacAIShaped-v3').env
# env = gym.wrappers.RecordEpisodeStatistics(env)
# env.action_space.seed(0)
obs = env.reset()
env.render()
except Exception as e:
e.printStackTrace()
# print("reward is", env.step([[[ 17, 2 , 0 , 3 , 0 , 1 , 2, 0]]])[1])
# env.render()
# print("reward is", env.step([[[ 34 , 4 , 1 ,2 , 1 , 2 , 3 ,0]]])[1])
# env.render()
# print("reward is", env.step([[[ 14*16+14 , 1 , 0 ,0 , 0 , 0 , 0 ,0]]])[1])
save_code=True)
wandb.tensorboard.patch(save=False)
writer = SummaryWriter(f"/tmp/{experiment_name}")
CHECKPOINT_FREQUENCY = 50
# TRY NOT TO MODIFY: seeding
device = torch.device(
'cuda' if torch.cuda.is_available() and args.cuda else 'cpu')
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.backends.cudnn.deterministic = args.torch_deterministic
envs = MicroRTSGridModeVecEnv(
num_selfplay_envs=args.num_selfplay_envs,
num_bot_envs=args.num_bot_envs,
max_steps=2000,
render_theme=2,
ai2s=[microrts_ai.coacAI for _ in range(args.num_bot_envs)],
map_path="maps/16x16/basesWorkers16x16.xml",
reward_weight=np.array([10.0, 1.0, 1.0, 0.2, 1.0, 4.0]))
envs = MicroRTSStatsRecorder(envs, args.gamma)
envs = VecMonitor(envs)
if args.capture_video:
envs = VecVideoRecorder(envs,
f'videos/{experiment_name}',
record_video_trigger=lambda x: x % 1000000 == 0,
video_length=2000)
# if args.prod_mode:
# envs = VecPyTorch(
# SubprocVecEnv([make_env(args.gym_id, args.seed+i, i) for i in range(args.num_envs)], "fork"),
# device
# )
"coacAI": microrts_ai.coacAI,
"naiveMCTSAI": microrts_ai.naiveMCTSAI,
"mixedBot": microrts_ai.mixedBot,
"rojo": microrts_ai.rojo,
"izanagi": microrts_ai.izanagi,
"tiamat": microrts_ai.tiamat,
"droplet": microrts_ai.droplet,
"guidedRojoA3N": microrts_ai.guidedRojoA3N
}
ai_names, ais = list(all_ais.keys()), list(all_ais.values())
ai_match_stats = dict(zip(ai_names, np.zeros((len(ais), 3))))
args.num_envs = len(ais)
envs = MicroRTSGridModeVecEnv(num_bot_envs=len(ais),
num_selfplay_envs=0,
render_theme=2,
ai2s=ais,
map_path="maps/16x16/basesWorkers16x16A.xml",
reward_weight=np.array(
[10.0, 1.0, 1.0, 0.2, 1.0, 4.0]))
envs = MicroRTSStatsRecorder(envs)
envs = VecMonitor(envs)
if args.capture_video:
envs = VecVideoRecorder(envs,
f'videos/{experiment_name}',
record_video_trigger=lambda x: x % 1000000 == 0,
video_length=2000)
assert isinstance(
envs.action_space,
MultiDiscrete), "only MultiDiscrete action space is supported"