if args.obj is not None:
obj = args.obj
envs = []
for g, seeds in zip(graphs, e_seeds_list):
env = NetworkEnv(fullGraph=g,
seeds=seeds,
opt_reward=0,
nop_r=args.nop_reward,
times_mean=args.times_mean_env,
bad_reward=args.bad_reward,
clip_max=args.max_reward,
clip_min=args.min_reward,
normalize=args.norm_reward)
envs.append(env)
replay = PriortizedReplay(BUFF_SIZE, 10, beta=0.6)
# BUFF_SIZE2 = 80
replay_her = PriortizedReplay(BUFF_SIZE, 10, beta=0.6)
goal_envs = []
for g, seeds in zip(graphs, e_seeds_list):
env = NetworkEnv2(fullGraph=g,
seeds=seeds,
opt_reward=0,
nop_r=args.nop_reward,
times_mean=2,
bad_reward=args.bad_reward,
clip_max=args.max_reward,
clip_min=args.min_reward,
normalize=args.norm_reward,
budget=budget)
ch.append(np.mean(rs))
print("Change for %s is %f" % (gp, ch[-1]))
logging.info('Change Results:' + str(obj1) + ' ' + str(S1))
final_goal = 0
if args.obj is not None:
obj = args.obj
envs = []
for g, seeds in zip(graphs, e_seeds_list):
env = NetworkEnv(fullGraph=g, seeds=seeds, opt_reward=0, nop_r=args.nop_reward,
times_mean=args.times_mean_env, bad_reward=args.bad_reward, clip_max=args.max_reward,
clip_min=args.min_reward, normalize=args.norm_reward)
envs.append(env)
replay = PriortizedReplay(BUFF_SIZE, 10, beta=0.6)
# BUFF_SIZE2 = 80
replay_her = PriortizedReplay(BUFF_SIZE, 10, beta=0.6)
goal_envs = []
for g, seeds in zip(graphs, e_seeds_list):
env = NetworkEnv2(fullGraph=g, seeds=seeds, opt_reward=0, nop_r=args.nop_reward,
times_mean=args.times_mean_env, bad_reward=args.bad_reward, clip_max=args.max_reward,
clip_min=args.min_reward, normalize=args.norm_reward, budget=budget)
goal_envs.append(env)
logging.info('State Dimensions: ' + str(action_dim))
logging.info('Action Dimensions: ' + str(action_dim))
acmodel = DQNTrainer(input_dim=input_dim, state_dim=action_dim, action_dim=action_dim, replayBuff=replay, lr=LR,
use_cuda=use_cuda, gamma=args.gamma,