ep = 10000
tau = 0.0001
gamma = 0.99
min_batch = 32
actor_lr = 0.00001
critic_lr = 0.0001
buffer_size = 1000000
layers = [1024, 512]
state_dim = env.observation_space["observation"].shape[
0] + env.observation_space["desired_goal"].shape[0]
action_dim = env.action_space.shape[0]
action_bound = env.action_space.high
actor_noise = OUNoise(mu=np.zeros(action_dim))
actor = ActorNetwork(sess, state_dim, action_dim, action_bound, layers,
actor_lr, tau, min_batch)
critic = CriticNetwork(sess, state_dim, action_dim, layers, critic_lr,
tau, gamma, actor.get_num_trainable_vars())
tf.compat.v1.summary.FileWriter("logdir/graphpend",
graph=tf.compat.v1.get_default_graph())
print("\033[0;1;32m")
print("===================")
print("LE DEBUT")
print("===================")
if "--demo" in sys.argv:
if "--load" in sys.argv:
print("loading weights")
loader = tf.compat.v1.train.Saver()
arg_index = sys.argv.index("--load")
def main():
with tf.compat.v1.Session() as sess:
tic = time.time()
env = customEnv()
if "--mstep" in sys.argv:
arg_index = sys.argv.index("--mstep")
micro_stepping = int(sys.argv[arg_index + 1])
else:
micro_stepping = 1
if "--ep" in sys.argv:
arg_index = sys.argv.index("--ep")
ep = int(sys.argv[arg_index + 1])
else:
ep = 10000
tau = 0.001
gamma = 0.99
min_batch = 64
actor_lr = 0.0001
critic_lr = 0.001
buffer_size = 1000000
layers = [300]
state_dim = (env.observation_space["observation"].shape[0] + env.observation_space["desired_goal"].shape[0])*micro_stepping
action_dim = env.action_space.shape[0]
action_bound = env.action_space.high
actor = ActorNetwork(sess, state_dim, action_dim, action_bound, layers, actor_lr, tau, min_batch)
critic = CriticNetwork(sess, state_dim, action_dim, layers, critic_lr, tau, gamma, actor.get_num_trainable_vars())
action_wanted = tf.compat.v1.placeholder(tf.float32, (None, action_dim))
reward_wanted = tf.compat.v1.placeholder(tf.float32, (None, 1))
actor_target = tf.reduce_mean(tf.square(actor.out-action_wanted))
critic_target = tf.reduce_mean(tf.square(critic.out-reward_wanted))
actor_train = tf.compat.v1.train.AdamOptimizer(actor_lr).minimize(actor_target)
critic_train = tf.compat.v1.train.AdamOptimizer(critic_lr).minimize(critic_target)
update_target_network_actor = [actor.target_network_params[i].assign(actor.network_params[i]) for i in range(len(actor.target_network_params))]
update_target_network_critic = [critic.target_network_params[i].assign(critic.network_params[i]) for i in range(len(critic.target_network_params))]
print("\033[0;1;32m")
print("===================")
print("LE DEBUT")
print("===================")
print("loading buffer")
arg_index = sys.argv.index("--loadBuff")
buffPath = sys.argv[arg_index + 1]
buffer = compress_json.local_load("preTrain/"+buffPath+".json.gz")
print("buffer loaded")
sess.run(tf.compat.v1.global_variables_initializer())
saver = tf.compat.v1.train.Saver()
i = 0
while i < ep:
i += 1
states, actions, rewards = sample(buffer,min_batch)
sess.run(actor_train,{actor.inputs: states, action_wanted: actions})
sess.run(critic_train,{critic.inputs: states, critic.action: actions, reward_wanted: np.reshape(rewards,(min_batch,1))})
print("\033[0;1;4;97m", end='')
print("miniBatch {} / {}".format(i,ep), end='')
print("\033[0;m ", end='')
tac = time.time()
print("\033[3;91m", end='')
print("{} secondes".format(int(tac - tic)), end='')
print("\033[0;m \r", end='')
sess.run(update_target_network_actor)
sess.run(update_target_network_critic)
arg_index = sys.argv.index("--save")
save_name = sys.argv[arg_index + 1]
saver.save(sess, "savedir/" + save_name+"/save")
print("\033[0;1;32m")
print("session saved at : " + save_name)
return 0
ep = 10000
tau = 0.001
gamma = 0.99
min_batch = 64
actor_lr = 0.0001
critic_lr = 0.001
buffer_size = 1000000
layers = [400,300]
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
action_bound = env.action_space.high
actor_noise = OUNoise(mu=np.zeros(action_dim))
actor = ActorNetwork(sess, state_dim, action_dim, action_bound, layers, actor_lr, tau, min_batch)
critic = CriticNetwork(sess, state_dim, action_dim, layers, critic_lr, tau, gamma, actor.get_num_trainable_vars())
tf.compat.v1.summary.FileWriter("logdir/graphpend", graph=tf.compat.v1.get_default_graph())
print("\033[0;1;32m")
print("===================")
print("LE DEBUT")
print("===================")
if "--demo" in sys.argv:
if "--load" in sys.argv:
print("loading weights")
loader = tf.compat.v1.train.Saver()
arg_index = sys.argv.index("--load")
save_name = sys.argv[arg_index + 1]
loader.restore(sess,"savedir/"+save_name+"/save")
print("weights loaded")