import numpy as np
import DQN
NUM_EPISODE = 10
NUM_STEP = 100
if __name__ == "__main__":
mainQN = DQN.QNetwork(debug_log=True)
memory = DQN.Memory(max_size=1000)
#actor = DQN.Actor()
for episode in range(NUM_EPISODE):
print('episode {}'.format(episode))
state = np.random.rand(16 * 16 * 8).reshape(1, 16, 16, 8)
#action1 = [7, 7]
#action2 = [8, 8]
for step in range(NUM_STEP):
#action, _ = actor.get_action(state, step, mainQN, 'r', action1, action2, 1, True, False, False)
action = np.array([0, 0])
if step == NUM_STEP - 1:
next_state = np.zeros((1, 16, 16, 8))
reward = 1.0
else:
next_state = np.random.rand(16 * 16 * 8).reshape(1, 16, 16, 8)
solved_reward = 230 # stop training if avg_reward > solved_reward
log_interval = 20 # print avg reward in the interval
max_episodes = 50000 # max training episodes
max_timesteps = 3000 # max timesteps in one episode
n_latent_var = 64 # number of variables in hidden layer
update_timestep = 200 # update policy every n timesteps
#Change these first
lr = 0.002
betas = (0.9, 0.999)
gamma = 0.99 # discount factor
K_epochs = 4 # update policy for K epochs
eps_clip = 0.2 # clip parameter for PPO
#############################################
# print(dir(DQN))
memory = DQN.Memory()
model = DQN.DQN(state_dim, action_dim, n_latent_var, lr, betas, gamma,
K_epochs, eps_clip)
# memory = PPO.Memory()
# model = PPO.PPO(state_dim, action_dim, n_latent_var, lr, betas, gamma, K_epochs, eps_clip)
print("About to load model...")
if loadingBool:
try:
print(file)
## DQN
model.policy_net.load_state_dict(torch.load(file))
model.target_net.load_state_dict(torch.load(file))
## PPO
# model.policy.load_state_dict(torch.load(file))
