import gym
from ac import Agent
import numpy as np
from monitor import Monitor
from trainer import Trainer
from smoothie import Smoothie
agent = Agent(0.00001, 0.00005)
env = gym.make("LunarLander-v2")
score_history = []
num_episodes = 10000
monitor = Monitor("monitor.csv",
title = title,
metadata = dict(
gamma=0.99,
agent_class=Agent.__name__,
pretrain=0,
copies=1,
report_interval=1,
comment = "Single agent training",
environment = env.__class__.__name__
),
plots=[
[
{
"label": "min train score",
"line_width": 1.0
},
{
"label": "train score"
},
import gym
from ac import Agent
import numpy as np
from monitor import Monitor
agent = Agent(0.00001, 0.00005)
env = gym.make("LunarLander-v2")
score_history = []
num_episodes = 2000
monitor = Monitor("monitor.csv")
monitor.start_server(8080)
for t in range(num_episodes):
done = False
observation = env.reset()
score = 0.0
while not done:
action = agent.choose_action(observation)
observation_, reward, done, info = env.step(action)
agent.learn(observation, action, reward, observation_, done)
observation = observation_
score += reward
score_history.append(score)
avg_score = np.mean(score_history[-100:])
print("Episode:", t, " score:", score, " average score:", avg_score)
monitor.add(t, average_score = avg_score)
import numpy as np
import getopt, sys
from envs import make_env
from ac import ACAgent as Agent
np.set_printoptions(precision=3)
opts, args = getopt.getopt(sys.argv[1:], "vn:l:")
opts = dict(opts)
num_tests = int(opts.get("-n", 100))
do_render = "-v" in opts
load_from = opts.get("-l")
env = make_env(args[0])
num_actions = env.action_space.n
observation_shape = env.observation_space.shape
assert len(observation_shape) == 1
observation_dim = observation_shape[0]
agent = Agent(observation_dim, num_actions, 0.00001, 0.00005)
if load_from:
agent.load(load_from)
print("\n\nAgent weights loaded from:", load_from, "\n\n")
for t in range(num_tests):
score, _ = agent.run_episode(env, learn=True, render=do_render)
print(score)
env_name = args[0]
opts = dict(opts)
report_interval = int(opts.get("-r", 1))
test_interval = int(opts.get("-t", 100))
render = "-v" in opts
gamma = float(opts.get("-g", 0.99))
comment = opts.get("-c", "")
env = make_env(env_name)
num_actions = env.action_space.n
observation_shape = env.observation_space.shape
assert len(observation_shape) == 1
observation_dim = observation_shape[0]
agent = Agent(observation_dim, num_actions, 0.00001, 0.00005, gamma=gamma)
title = opts.get(
"-T", "Training agent %s in %s" % (agent.__class__.__name__, env_name))
score_history = []
num_episodes = 10000
monitor = Monitor("monitor.csv",
title=title,
metadata=dict(gamma=0.99,
agent_class=Agent.__name__,
pretrain=0,
copies=1,
report_interval=1,
comment=comment,
environment=env.__class__.__name__),