def setup_experiment(model_class, exp_type, seed):
if model_class not in (QLearning, Sarsa, ExpectedSarsa):
raise ValueError("Unknown model : %s" % (model_class, ))
if model_class == QLearning:
learning_rate = 0.85
discount_factor = 0.85
else:
learning_rate = 0.15
discount_factor = 0.85
epsilon = 0.15
train_episodes = 1000000
value_function = DictActionValueFunction(0)
learning_rate = StaticLearningRate(learning_rate)
discount_factor = StaticDiscountFactor(discount_factor)
snake_params = SnakeParameters()
env = SnakeEnvironment(snake_params)
policy = EpsilonGreedyPolicy(env, epsilon)
experiment = Experiment(env=env,
model_class=model_class,
experiment_type=exp_type,
seed=seed,
epsilon=epsilon,
train_episodes=train_episodes,
value_function=value_function,
learning_rate=learning_rate,
discount_factor=discount_factor,
policy=policy)
experiment.load()
return experiment
def run_sarsa():
params = SnakeParameters()
env = SnakeEnvironment(params)
value_function = pickle.load(
open("../models/sarsa_%s.p" % params.file_str, "rb"))
agent = SnakeAgent(policy=EpsilonGreedyPolicy(env, 0),
action_value_function=value_function)
start_app(env, agent, params)
def train():
for params in get_snake_parameters():
env = SnakeEnvironment(params)
params.policy = EpsilonGreedyPolicy(env, params.epsilon)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = Sarsa
exp_params.model_params = params
exp_params.seed = get_parameters_seed()
output_dir = "../../../models/sarsa/params/%i" % exp_params.seed
exp_params.model_output_dir = output_dir
train_and_store_model(exp_params)
def train_sarsa():
start = timer()
params = SnakeParameters()
env = SnakeEnvironment(params)
policy = EpsilonGreedyPolicy(env, params.epsilon)
learner = Sarsa(action_value_function=params.value_function,
policy=policy,
learning_rate=params.learning_rate,
discount_factor=params.discount_factor)
learner.train(env, params.train_episodes)
pickle.dump(learner.Q, open("../models/sarsa_%s.p" % params.file_str,
"wb"))
print("Elapsed time:", timer() - start)
def test_state():
params = SnakeParameters()
env = SnakeEnvironment(params)
state = BoardState(env)
print(state)
print(hash(state))
state = DirectionalState(env)
print(state)
print(hash(state))
state = SnakeFoodState(env)
print(state)
print(hash(state))
state = DirectionalDistanceState(env)
print(state)
print(hash(state))
def analyze_test():
params = SnakeParameters()
params.discount_factor = StaticDiscountFactor(0.95)
params.learning_rate = StaticLearningRate(0.15)
env = SnakeEnvironment(params)
params.policy = GreedyPolicy(env)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = Sarsa
exp_params.model_params = params
model_output_dir = "../../../models/sarsa/reward/%i" % exp_params.seed
exp_params.model_output_dir = model_output_dir
image_output_dir = "../../../images/sarsa/reward/%i" % exp_params.seed
exp_params.image_output_dir = image_output_dir
analyze_models_test(exp_params)
def train():
params = SnakeParameters()
params.discount_factor = StaticDiscountFactor(0.95)
params.learning_rate = StaticLearningRate(0.15)
params.reward = get_state_reward()
env = SnakeEnvironment(params)
params.policy = EpsilonGreedyPolicy(env, params.epsilon)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = Sarsa
exp_params.model_params = params
for seed in get_state_seeds():
exp_params.seed = seed
output_dir = "../../../models/sarsa/state/%i" % exp_params.seed
exp_params.model_output_dir = output_dir
train_models(exp_params)
def train():
params = SnakeParameters()
params.discount_factor = StaticDiscountFactor(0.85)
params.learning_rate = StaticLearningRate(0.85)
env = SnakeEnvironment(params)
params.policy = EpsilonGreedyPolicy(env, params.epsilon)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = QLearning
exp_params.model_params = params
for seed in get_reward_seeds():
exp_params.seed = seed
output_dir = "../../../models/qlearning/reward/%i" % exp_params.seed
exp_params.model_output_dir = output_dir
for state in get_reward_states():
exp_params.model_params.state = state
train_models(exp_params)
def main():
params = SnakeParameters()
env = SnakeEnvironment(params)
player = SnakePlayer()
start_app(env, player, params)
@pyqtSlot()
def __on_push_run(self):
print("Run Agent")
@pyqtSlot()
def __on_push_train(self):
print("Train Agent")
@pyqtSlot()
def __on_push_reward_stat(self):
print("Reward Stat")
@pyqtSlot()
def __on_push_action_stat(self):
print("Action Stat")
@pyqtSlot()
def __on_push_exploration_exploitation_stat(self):
print("Exploration/Exploitation Stat")
if __name__ == '__main__':
params = SnakeParameters()
env = SnakeEnvironment(params)
player = SnakePlayer()
app = QApplication(["SnakeBot"])
window = Window(env, player, params)
window.show()
app.exec_()