def main():
actor_learning_rate = [
1 * 10**-4, 3 * 10**-4, 6 * 10**-4, 10**-3, 3 * 10**-3, 6 * 10**-3,
10**-2
]
critic_learning_rate = [
1 * 10**-4, 3 * 10**-4, 6 * 10**-4, 10**-3, 3 * 10**-3, 6 * 10**-3,
10**-2
]
tau = [.9, .93, .95, .97, .99]
batch_size = [32, 64, 128, 256]
p_rand = [0, .1, .2, .3, .4]
sigma = [0, .1, .2, .3, .4]
L2_norm_coeff = [0, .01, .03, .1, .3, .6, 1]
load_checkpoint = False
env = gym.make("CartPoleContinuousBulletEnv-v0")
agent = Agent(input_dims=env.observation_space.shape,
n_actions=env.action_space.shape[0])
episodes = 250
filename = 'MoutainCarContinuous.png'
figure_file = 'plots/' + filename
best_score = env.reward_range[0]
score_history = []
if load_checkpoint:
agent.load_models()
env.render(mode='human')
for i in range(episodes):
observation = env.reset()
done = False
score = 0
while not done:
action = agent.choose_action(observation)
observation_, reward, done, info = env.step(action)
score += reward
agent.remember(observation, action, reward, observation_, done)
if not load_checkpoint:
agent.learn()
else:
env.render()
observation = observation_
score_history.append(score)
avg_score = np.mean(score_history[-100:])
if avg_score > best_score:
best_score = avg_score
if not load_checkpoint:
agent.save_models()
print("episode", i, "score", score, "average score", avg_score)
if not load_checkpoint:
x = [i + 1 for i in range(episodes)]
plot_error_drop(x, score_history)
from ddpg_torch import Agent
import gym
import numpy as np
#from utils import PlotLearning
env = gym.make('LunarLanderContinuous-v2')
agent = Agent(alpha=0.000025,beta=0.00025, input_dims=[8],tau=0.001,env=env)
np.random.seed(0)
agent.load_models()
score_history = []
ep = 0
while True:
ep += 1
done = False
score = 0
obs = env.reset()
while not done:
env.render()
act = agent.choose_action(obs)
new_state,reward,done,info = env.step(act)
agent.remember(obs,act,reward,new_state,int(done))
#agent.learn()
score += reward
obs = new_state
score_history.append(score)
print('episode',ep,'score%.2f' % score, '100 game average %.2f' % np.mean(score_history[-100:]))
if ep % 25 == 0:
agent.save_models()
if len(score_history) >= 25 and np.mean(score_history[-25:]) > 200:
break
filename = 'pendulum.png'
#plotLearning(score_history, filename, window=100)
str(agent.beta) + '_' + str(n_games) + '_games'
figure_file = 'plots/' + filename + '.png'
if load_checkpoint:
agent.load_models()
best_score = env.reward_range[0]
score_history = []
for i in range(n_games):
observation = env.reset()
observation=observation.reshape(observation.shape[0]*observation.shape[1],)
done = False
score = 0
agent.noise.reset()
while not done:
action = agent.choose_action(observation)
observation_, reward, done, info = env.step(action)
observation_ = observation_.reshape(observation_.shape[0]*observation_.shape[1],)
if not test_agent:
agent.remember(observation, action, reward, observation_, done)
agent.learn()
score += reward
observation = observation_
env.render()
score_history.append(score)
avg_score = np.mean(score_history[-100:])
if not test_agent:
if avg_score > best_score:
best_score = avg_score