'epsi_high': 0.9,
'epsi_low': 0.05,
'decay': 200,
'lr': 0.001,
'capacity': 10000,
'batch_size': 32,
'state_space_dim': env.state_dim,
'action_space_dim': env.action_dim,
}
agent = Agent(**params)
score = []
mean = []
for episode in range(1000):
s0 = env.reset()
total_reward = 1
while True:
env.render()
a0 = agent.act(s0)
s1, r1, done= env.step(a0)
if done:
r1 = -1
agent.put(s0, a0, r1, s1)
if done:
break
total_reward += r1
# Gloabel Variable
MAX_EPISOSES = 500
MAX_EP_STEPS = 500
# Set the environement
env = ArmEnv()
s_dim = env.state_dim
a_dim = env.action_dim
a_bound = env.action_bound
# set the RL method
rl = DDPG(a_dim, s_dim, a_bound)
# start Training
for i in range(MAX_EPISOSES):
s = env.reset()
for j in range(MAX_EP_STEPS):
env.render()
a = rl.choose_action(s)
s_, r, done = env.step(a)
rl.store_transitions(s, a, r, s_)
if rl.memory_full:
# start to learn once has fulfulled the memory
rl.learn()
s = s_
def eval():
rl.restore()
env.render()
env.viewer.set_vsync(True)
while True:
s = env.reset()
for _ in range(200):
env.render()
a = rl.choose_action(s)
s, r, done = env.step(a)
if done:
break
if ON_TRAIN:
train()
else:
eval()
# summary:
"""
env should have at least:
env.reset()
env.render()
env.step()
while RL should have at least:
rl.choose_action()
rl.store_transition()
rl.learn()
rl.memory_full
"""
MAX_EPISODES = 500
MAX_EP_STEPS = 200
# set env
env = ArmEnv()
s_dim = env.state_dim
a_dim = env.action_dim
a_bound = env.action_bound
# set RL method
rl = DDPG(a_dim, s_dim, a_bound)
# start training
for i in range(MAX_EPISODES):
s = env.reset()
for j in range(MAX_EP_STEPS):
env.render()
a = rl.choose_action(s)
s_, r, done = env.step(a)
rl.store_transition(s, a, r, s_)
if rl.memory_full:
# start to learn once has fulfilled the memory
rl.learn()
s = s_