def training_4():
snakes_games = [SnakeAI([30, 30]), SnakeAIBorders([30, 30])]
for snake in snakes_games:
rewards = [
{
'aprox': [1, 0],
'eat': [10, 0],
'dead': [-100, 0]
},
{
'aprox': [2, -1],
'eat': [10, 0],
'dead': [-100, 0]
},
]
for reward in rewards:
agent_environment = SnakeGameEnv(snake, reward,
len(snakes_games[0].state()))
for i in range(3):
Lp = LearningProcess(agent_environment)
Lp.pre_learning_process()
returns, losses = Lp.training()
visualization(Lp, returns, losses)
path = Lp.policy_saver(i)
save_returns = path + '/returns.csv'
save_losses = path + '/losses.csv'
write_data(save_returns, returns)
write_data(save_losses, losses)
def playing_AI():
snake = SnakeAI([30, 30])
agent_environment = SnakeGameEnv(snake)
screen = Screen(300, 300, [30, 30], 5)
Lp = LearningProcess(agent_environment)
Lp.play_previous_policy(screen)
def training():
snake = SnakeAI([30, 30])
agent_environment = SnakeGameEnv(snake, -100)
Lp = LearningProcess(agent_environment)
Lp.pre_learning_process()
Lp.training()
Lp.policy_saver()
def samples_1():
snake = SnakeAI([30, 30])
rewards = {'aprox': [1, 0], 'eat': [10, 0], 'dead': [-100, 0]}
agent_environment = SnakeGameEnv(snake, rewards, 33)
Lp = LearningProcess(agent_environment)
policy = Lp.load_previous_policy(0, 0)
return compute_avg_return(Lp.eval_env, policy, 30)
def training_2():
snake = SnakeAIBorders([30, 30])
rewards = {'aprox': [1, 0], 'eat': [10, 0], 'dead': [-100, 0]}
agent_environment = SnakeGameEnv(snake, rewards, 33)
Lp = LearningProcess(agent_environment)
Lp.pre_learning_process()
Lp.training()
Lp.policy_saver()
def training_3():
snake = SnakeAIBorders([30, 30])
rewards = {'aprox': [1, 0], 'eat': [10, 0], 'dead': [-100, 0]}
agent_environment = SnakeGameEnv(snake, rewards, 33)
Lp = LearningProcess(agent_environment)
Lp.pre_learning_process()
returns, losses = Lp.training()
visualization(Lp, returns, losses)
path = Lp.policy_saver()
save_returns = path + '/returns.csv'
save_losses = path + '/losses.csv'
write_data(save_returns, returns)
write_data(save_losses, losses)
returns_read = read_data(save_returns)
losses_read = read_data(save_losses)
visualization(Lp, returns_read, losses_read)
def training_loops_2(snakes_games, rewards, redundance):
for snake in snakes_games:
for reward in rewards:
agent_environment = SnakeGameEnv(snake, reward, len(snake.state()))
Lp = LearningProcess(agent_environment)
for i in range(redundance):
print('*-*' * 15)
print(
'iteration {} using the reward {} with the rules/game {} and input {}'
.format(i, reward, snake.__class__.__name__,
len(snake.state())))
Lp.pre_learning_process()
returns, losses = Lp.training()
path = Lp.policy_saver(i)
save_returns = path + '/returns.csv'
save_losses = path + '/losses.csv'
write_data(save_returns, returns)
write_data(save_losses, losses)
def sampler(snake, reward, num_reward=0, iteration=0, num_episodes=30, points=False):
"""Evaluates n samples of a trained network.
snake (object): type of game.
reward (dictionary): reward of the snake game environment.
num_reward (int): Number wich indentifies the reward.
iteration (int): Number of the iteration.
num_episodes (int): amount of samples to be taken.
points (bool): If its true then the return will be the a list of lists wich
contains the amount of points per step. The lists have a length
accord with the number of steps the snake 'survived'.
"""
agent_environment = SnakeGameEnv(snake, reward, len(snake.state()))
Lp = LearningProcess(agent_environment)
policy = Lp.load_previous_policy(num_reward, iteration)
if points:
return points_history(Lp.sample_env, policy, num_episodes)
else:
return compute_avg_return(Lp.sample_env, policy, num_episodes)
def training_loops(snakes_games, rewards, redundance):
"""Total training loop for multiple types of games and rewards, each type of game
will be proved with all the rewards an n number of iterations.
snake_games (list): List of objects to be train.
rewards (list): List of dictionaries (rewards of the snake game).
redundance (int): Number of iterations.
"""
for snake in snakes_games:
for num, reward in enumerate(rewards):
agent_environment = SnakeGameEnv(snake, reward, len(snake.state()))
Lp = LearningProcess(agent_environment)
for i in range(redundance):
print('*-*'*15)
print('iteration {} using the reward {} with the rules/game {} and input {}'.format(i, reward, snake.__class__.__name__, len(snake.state())))
Lp.pre_learning_process()
returns, losses = Lp.training()
path = Lp.policy_saver(num, i)
save_returns = path + '/returns.csv'
save_losses = path + '/losses.csv'
write_data(save_returns, returns)
write_data(save_losses, losses)
def samples_3(snake, reward, num_reward=0, iteration=0, num_episodes=30):
agent_environment = SnakeGameEnv(snake, reward, len(snake.state()))
Lp = LearningProcess(agent_environment)
policy = Lp.load_previous_policy(num_reward, iteration)
return points_history(Lp.sample_env, policy, num_episodes)
def samples_2(snake, reward, num_reward=0, iteration=0, num_episodes=30):
agent_environment = SnakeGameEnv(snake, reward, len(snake.state()))
Lp = LearningProcess(agent_environment)
policy = Lp.load_previous_policy(num_reward, iteration)
return compute_avg_return(Lp.eval_env, policy, num_episodes)