TOTAL_GAMETIME = 10000
N_EPISODES = 10000
REPLACE_TARGET = 10
game = GameEnv.RacingEnv()
game.fps = 60
GameTime = 0
GameHistory = []
renderFlag = True
ddqn_agent = DDQNAgent(alpha=0.0005,
gamma=0.99,
n_actions=5,
epsilon=0.02,
epsilon_end=0.01,
epsilon_dec=0.999,
replace_target=REPLACE_TARGET,
batch_size=64,
input_dims=19,
fname='ddqn_model.h5')
ddqn_agent.load_model()
ddqn_agent.update_network_parameters()
ddqn_scores = []
eps_history = []
def run():
#scores = deque(maxlen=100)
import gym
from gym import wrappers
import numpy as np
from ddqn_keras import DDQNAgent
from utils import plotLearning
if __name__ == '__main__':
env = gym.make('LunarLander-v2')
ddqn_agent = DDQNAgent(alpha=0.0005,
gamma=0.99,
n_actions=4,
epsilon=1.0,
batch_size=64,
input_dims=8)
n_games = 500
#ddqn_agent.load_model()
ddqn_scores = []
eps_history = []
#env = wrappers.Monitor(env, "tmp/lunar-lander-ddqn-2",
# video_callable=lambda episode_id: True, force=True)
for i in range(n_games):
done = False
score = 0
observation = env.reset()
while not done:
action = ddqn_agent.choose_action(observation)
observation_, reward, done, info = env.step(action)
score += reward
ddqn_agent.remember(observation, action, reward, observation_,
int(done))
from ddqn_keras import DDQNAgent
from collections import deque
import random, math
TOTAL_GAMETIME = 1000 # Max game time for one episode
N_EPISODES = 10000
REPLACE_TARGET = 50
game = GameEnv.RacingEnv()
game.fps = 60
GameTime = 0
GameHistory = []
renderFlag = False
ddqn_agent = DDQNAgent(alpha=0.0005, gamma=0.99, n_actions=5, epsilon=1.00, epsilon_end=0.10, epsilon_dec=0.9995, replace_target= REPLACE_TARGET, batch_size=512, input_dims=19)
# if you want to load the existing model uncomment this line.
# careful an existing model might be overwritten
#ddqn_agent.load_model()
ddqn_scores = []
eps_history = []
def run():
for e in range(N_EPISODES):
game.reset() #reset env
done = False
import gym
from gym import wrappers
import numpy as np
import random
from ddqn_keras import DDQNAgent
from utils import plotLearning
import ipdb
SHOW_EVERY = 10
if __name__ == '__main__':
env = gym.make('MountainCar-v0')
# ipdb.set_trace()
ddqn_agent = DDQNAgent(alpha=0.0005,
gamma=0.99,
n_actions=env.action_space.n,
epsilon=.01,
batch_size=64,
input_dims=env.observation_space.shape[0])
n_games = 500
#ddqn_agent.load_model()
ddqn_scores = []
eps_history = []
#env = wrappers.Monitor(env, "tmp/lunar-lander-ddqn-2",
# video_callable=lambda episode_id: True, force=True)
for i in range(n_games):
done = False
score = 0
observation = env.reset()
while not done:
if not i or not i % SHOW_EVERY:
def close(self):
pygame.quit()
# Clock to limit speed
clock = pygame.time.Clock()
# Exit the program?
exit_program = False
env = Arcanoid()
ddqn_agent = DDQNAgent(alpha=0.005,
gamma=0.99,
n_actions=env.action_space.n,
epsilon=1.0,
batch_size=64,
input_dims=env.observation_space.shape[0] * 2,
replace_target=1000)
ddqn_scores = []
eps_history = []
history = []
n_games = 500_000
start = time.time()
for i in range(n_games):
done = False
observation = env.reset()
observation = observation / [GAME_WIDTH, GAME_WIDTH, GAME_HEIGHT]
observation = np.append([-1, -1, -1], observation)
while not done:
action = ddqn_agent.choose_action(observation)