def __init__(self, config_file_name, seed=0, verbose=False):
self.verbose = verbose
self.env = blackbox.EnvBlackBox(seed)
#print environment info
if (verbose):
self.env.print_info()
#init DQN agent
self.agent = agent_dqn.DQNAgent(self.env, config_file_name, 0.3, 0.05,
0.99999)
#self.agent = agent.Agent(self.env)
#iterations count
self.training_iterations = 100000
self.testing_iterations = 10000
import sys
sys.path.append("..") # Adds higher directory to python modules path.
import libs.libs_env.env_birds
import agent
import agent_dqn
env = libs.libs_env.env_birds.EnvBirds()
env.print_info()
#agent = agent.Agent(env)
agent = agent_dqn.DQNAgent(env, "flappy_bird_net.json")
training_iterations = 500000
for i in range(0, training_iterations):
agent.main()
if (i % 100) == 0:
progress = 100.0 * i / training_iterations
print("training done = ", progress, " score = ", env.get_score())
env.reset_score()
agent.run_best_enable()
testing_iterations = 100000
for i in range(0, testing_iterations):
agent.main()
#example for convolutional and deep neural network use
import sys
sys.path.append("..") # Adds higher directory to python modules path.
import libs.libs_env.env_pong
import agent_dqn
#init environment
env = libs.libs_env.env_pong.EnvPong()
#print environment info
env.print_info()
#init DQN agent
agent = agent_dqn.DQNAgent(env, "pong_network.json", 0.2, 0.01, 0.99999)
#process training
training_iterations = 200000
for iteration in range(0, training_iterations):
agent.main()
#print training progress %, ane score, every 100th iterations
if iteration % 100 == 0:
print(iteration * 100.0 / training_iterations, env.get_score())
#reset score
env.reset_score()
#choose only the best action
agent.run_best_enable()
env = libs.libs_env.env_atari_arkanoid.EnvAtariArkanoid()
#print environment info
env.print_info()
'''
#random play environment test
random_agent = agent.Agent(env)
while True:
random_agent.main()
if env.get_iterations()%256 == 0:
print(" miss ", env.get_miss(), " iterations = ", env.get_iterations())
env.render()
'''
#init DQN agent
dqn_agent = agent_dqn.DQNAgent(env, "atari_arkanoid_network.json", 0.4, 0.1,
0.99999)
#process training
total_games_to_play = 500
while env.get_games_count() < total_games_to_play:
dqn_agent.main()
#print training progress %, ane score, every 100th iterations
if env.get_iterations() % 256 == 0:
env._print()
env.render()
if env.get_iterations() % 256 == 0:
print("done = ",
env.get_games_count() * 100.0 / total_games_to_play, "%",
env = env_black_box.EnvBlackBox(4)
#print environment info
env.print_info()
#random play environment test
random_agent = agent.Agent(env)
while True:
random_agent.main()
if env.get_iterations() % 256 == 0:
print(" iterations = ", env.get_iterations(), " score = ",
env.get_score())
#init DQN agent
dqn_agent = agent_dqn.DQNAgent(env, "black_box_network.json", 0.1, 0.05,
0.99999)
#process training
training_iterations = 100000
for i in range(0, training_iterations):
dqn_agent.main()
if env.get_iterations() % 256 == 0:
print(" iterations = ", env.get_iterations(), " score = ",
env.get_score(), " epsilon = ", dqn_agent.get_epsilon_training())
#reset score
env.reset_score()
#choose only the best action
import sys
sys.path.append("..") # Adds higher directory to python modules path.
import libs.libs_env.env_pong
import libs.libs_env.env_arkanoid
import agent
import agent_dqn
#env = libs.libs_env.env_pong.EnvPong()
env = libs.libs_env.env_arkanoid.EnvArkanoid()
env.print_info()
#agent = agent.Agent(env)
agent = agent_dqn.DQNAgent(env, "arkanoid_network.json")
training_iterations = 250000
for i in range(0, training_iterations):
agent.main()
if (i % 100) == 0:
progress = 100.0 * i / training_iterations
print("training done = ", progress, " score = ", env.get_score())
agent.save("arkanoid_network/")
env.reset_score()
agent.run_best_enable()