def plot_average_actions_over_time(learner):
actions = learner.actions_per_episode
average_actions = compute_mean_over_time(actions)
plt.semilogx(average_actions)
plt.title("Average # of Actions Over Time")
plt.xlabel("# of episodes")
plt.ylabel("Avg. # of actions")
def plot_average_reward_over_time(learner):
rewards_over_time = learner.rewards_per_episode
mean_over_time = compute_mean_over_time(rewards_over_time)
plt.semilogx(mean_over_time)
plt.title("Average Reward Over Time")
plt.xlabel("# of episodes")
plt.ylabel("Avg. Reward")
def get_corr_coefs(models):
corr_coefs = []
for model in models:
average_reward_over_time = compute_mean_over_time(
model.rewards_per_episode)
average_food_count_over_time = compute_mean_over_time(
model.food_count_per_episode)
if np.all(average_reward_over_time == 0):
continue
corr_coef = np.corrcoef(average_reward_over_time,
average_food_count_over_time)
corr_coefs.append(corr_coef)
return corr_coefs
def diagnostic_plot_exploration_vs_exploitation_over_time(learner):
actions = learner.actions_per_episode
exploratory_action = learner.exploratory_actions_per_episode
exploration = compute_mean_over_time(exploratory_action / actions)
plt.semilogx(exploration)
plt.title("Exploration vs. Exploitation")
plt.xlabel("# of episodes")
plt.ylabel("Exploration (%)")
def plot_multi_average_game_score_over_time(x, game_score_per_episode, labels):
average_game_score_over_time = [
compute_mean_over_time(gspe) for gspe in game_score_per_episode
]
for label, agsot in zip(labels, average_game_score_over_time):
plt.semilogx(x, agsot, label=label)
plt.title("Average Game Score Over Time")
plt.xlabel("# of episodes")
plt.ylabel("Game Score")
plt.legend(fontsize='x-small')
def plot_multi_average_food_count_over_time(x, food_count_per_episode, labels):
average_food_count_over_time = [
compute_mean_over_time(fcpe) for fcpe in food_count_per_episode
]
for label, afcot in zip(labels, average_food_count_over_time):
plt.semilogx(x, afcot, label=label)
plt.title("Average Food Count Over Time")
plt.xlabel("# of episodes")
plt.ylabel("Food Count")
plt.legend(fontsize='x-small')
def plot_multi_average_actions_over_time(x, actions_per_episode, labels):
average_actions_over_time = [
compute_mean_over_time(ape) for ape in actions_per_episode
]
for label, aaot in zip(labels, average_actions_over_time):
plt.semilogx(x, aaot, label=label)
plt.title("Average # of Actions Over Time")
plt.xlabel("# of episodes")
plt.ylabel("# of actions")
plt.legend(fontsize='x-small')
def plot_multi_average_reward_over_time(x, rewards_per_episode, labels):
mean_reward_over_time = [
compute_mean_over_time(rpe) for rpe in rewards_per_episode
]
for label, mrot in zip(labels, mean_reward_over_time):
plt.semilogx(x, mrot, label=label)
plt.title("Average Reward Over Time")
plt.xlabel("# of episodes")
plt.ylabel("Reward")
plt.legend(fontsize='x-small')
def plot_multi_average_self_collision_death_over_time(
x, self_collision_death_per_episode, labels):
average_self_collision_death_over_time = [
compute_mean_over_time(scdpe)
for scdpe in self_collision_death_per_episode
]
for label, ascdot in zip(labels, average_self_collision_death_over_time):
plt.semilogx(x, ascdot, label=label)
plt.title("Average Self-Collision Death Over Time")
plt.xlabel("# of episodes")
plt.ylabel("Self-Collision Death")
plt.legend(fontsize='x-small')