import six
from six.moves import range
import matplotlib.pyplot as plt
def calculate_cum_reward(policy):
"""Calculate cumulative reward with respect to time.
Parameters
----------
policy: bandit object
The bandit algorithm you want to evaluate.
Return
---------
cum_reward: dict
The dict stores {history_id: cumulative reward} .
cum_n_actions: dict
The dict stores
{history_id: cumulative number of recommended actions}.
"""
cum_reward = {-1: 0.0}
cum_n_actions = {-1: 0}
for i in range(policy.history_storage.n_histories):
reward = policy.history_storage.get_history(i).rewards
cum_n_actions[i] = cum_n_actions[i - 1] + len(reward)
cum_reward[i] = cum_reward[i - 1] + sum(six.viewvalues(reward))
return cum_reward, cum_n_actions
def calculate_avg_reward(policy):
"""Calculate average reward with respect to time.
Parameters
----------
policy: bandit object
The bandit algorithm you want to evaluate.
Return
---------
avg_reward: dict
The dict stores {history_id: average reward} .
"""
cum_reward, cum_n_actions = calculate_cum_reward(policy)
avg_reward = {}
for i in range(policy.history_storage.n_histories):
avg_reward[i] = cum_reward[i] / cum_n_actions[i]
return avg_reward
def plot_avg_reward(policy):
"""Plot average reward with respect to time.
Parameters
----------
policy: bandit object
The bandit algorithm you want to evaluate.
"""
avg_reward = calculate_avg_reward(policy)
plt.plot(avg_reward.keys(), avg_reward.values(), 'r-',
label="average reward")
plt.xlabel('time')
plt.ylabel('avg reward')
plt.legend()
plt.title("Average Reward with respect to Time")
def plot_avg_regret(policy):
"""Plot average regret with respect to time.
Parameters
----------
policy: bandit object
The bandit algorithm you want to evaluate.
"""
avg_reward = calculate_avg_reward(policy)
points = sorted(six.viewitems(avg_reward), key=lambda x: x[0])
x, y = zip(*points)
plt.plot(x, [1 - reward for reward in y], 'r-', label="average regret")
plt.xlabel('time')
plt.ylabel('avg regret')
plt.legend()
plt.title("Average Regret with respect to Time")
