#init DQN agent
agent = libs.libs_agent.agent_dqn.DQNAgent(
env, "networks/settlers_network/parameters.json", 0.2, 0.1) #0.2, 0.1
#agent = libs_agent.agent.Agent(env)
#process training
training_iterations = 500000
for iteration in range(0, training_iterations):
agent.main()
#print training progress %, ane score, every 100th iterations
if iteration % 100 == 0:
env._print()
print(iteration * 100.0 / training_iterations, env.get_score())
agent.save("networks/settlers_network/trained/")
agent.load("networks/settlers_network/trained/")
#reset score
env.reset_score()
#choose only the best action
agent.run_best_enable()
#process testing iterations
testing_iterations = 10000
for iteration in range(0, testing_iterations):
agent.main()
print("move=", env.get_move(), " score=", env.get_score(),
" moves to win=", env.get_moves_to_win())
env.render()
if env.get_iterations()%256 == 0:
str_progress = str(env.get_iterations()) + " "
str_progress+= str(env.get_games_count()) + " "
str_progress+= str(agent.get_epsilon_training()) + " "
str_progress+= str(env.get_score()) + " "
str_progress+= "\n"
training_progress_log.put_string(str_progress)
print("done = ", env.get_games_count()*100.0/total_games_to_play, "%", " eps = ", agent.get_epsilon_training(), " iterations = ", env.get_iterations())
if env.get_iterations()%50000 == 0:
print("SAVING network")
agent.save(network_path + "trained/")
agent.save(network_path + "trained/")
agent.load(network_path + "trained/")
#reset score
env.reset_score()
#choose only the best action
agent.run_best_enable()
'''
#init DQN agent
agent = libs.libs_agent.agent_dqn.DQNAgent(
env, "networks/arkanoid_network_b/parameters.json", 0.2, 0.02, 0.99999)
#process training
training_iterations = 250000
for iteration in range(0, training_iterations):
agent.main()
#print training progress %, ane score, every 100th iterations
if iteration % 100 == 0:
env._print()
agent.save("networks/arkanoid_network_b/trained/")
agent.load("networks/arkanoid_network_b/trained/")
#reset score
env.reset_score()
#choose only the best action
agent.run_best_enable()
#process testing iterations
testing_iterations = 10000
for iteration in range(0, testing_iterations):
agent.main()
env._print()