def test_should_update_an_arm_with_feedback(self):
arm_names = ['arm1', 'arm2']
arm = [EpsilonGreedyArm(name, n=1, reward_sum=1) for name in arm_names]
egreedy = EGreedy('my_bandit', arm, epsilon=0.1)
feedback = Feedback('arm1', reward=1)
egreedy.update(feedback)
assert egreedy.arms_dict['arm1'].reward_sum == 2 and egreedy.arms_dict[
'arm1'].n == 2
def update(self, bandit_id: str, arm_id: str, reward: float, payload: str = None) -> str:
"""
Updates bandit strategy given the feedback
:param bandit_id: [str], bandit id
:param arm_id: [str], arm name
:param reward: [float], reward of the chosen arm name
:param context: [dict, default=None], context used when choose was called.
"""
bandit: OnlineBandit = self.bandits[bandit_id]
bandit.update(Feedback(arm_id, reward, payload))
def test_should_update_an_arm_with_feedback(self):
arm_names = ['arm1', 'arm2', 'arm3']
arms = [BetaArm(name, n=2, n_rewards=1) for name in arm_names]
beta_bandit = BetaThompsonBandit('my_bandit', arms)
feedback = Feedback('arm2', reward=1)
beta_bandit.update(feedback)
assert beta_bandit.arms_dict['arm2'].n_rewards == 2
assert beta_bandit.arms_dict['arm2'].n == 3
def test_should_update_an_arm_with_feedback(self):
arm_names = ['arm1', 'arm2']
arms = [GaussianArm(name, 0.0001, 0, 1) for name in arm_names]
gaussian_bandit = GaussianThompsonBandit('my_bandit', arms)
feedback = Feedback('arm2', reward=10)
gaussian_bandit.update(feedback)
assert gaussian_bandit.arms_dict['arm2'].reward_sum == 10
assert gaussian_bandit.arms_dict['arm2'].squared_reward_sum == 101
assert gaussian_bandit.arms_dict['arm2'].n == 1.0001
def test_should_run_for_several_iterations_and_sample_all_arms(self):
arm_names = ['arm1', 'arm2', 'arm3']
arm = [EpsilonGreedyArm(name, n=1, reward_sum=1) for name in arm_names]
egreedy = EGreedy('my_bandit', arm, epsilon=0.1)
num_steps = 100
sampled_arm_ids = set()
for _ in range(num_steps):
arm = egreedy.choose()
sampled_arm_ids.add(arm.arm_id)
reward = 1.0 if random.random() < 0.2 else 0.0
egreedy.update(Feedback(arm.arm_id, reward))
assert sampled_arm_ids == set(arm_names)
def test_should_reset_value_to_default(self):
arm_names = ['arm1', 'arm2', 'arm3']
arms = [BetaArm(name, n=2, n_rewards=1) for name in arm_names]
beta_bandit = BetaThompsonBandit('my_bandit', arms)
num_steps = 20
for _ in range(num_steps):
arm = beta_bandit.choose()
reward = 1.0 if random.random() < 0.2 else 0.0
beta_bandit.update(Feedback(arm_id=arm.arm_id, reward=reward))
beta_bandit.reset()
assert all([arm.n_rewards == 1 and arm.n == 2 for arm in beta_bandit.arms_dict.values()])
def test_should_reset_value_to_default(self):
arm_names = ['arm1', 'arm2', 'arm3']
arms = [GaussianArm(name, 2, 2, 2) for name in arm_names]
gaussian_bandit = GaussianThompsonBandit('my_bandit', arms)
num_steps = 20
for _ in range(num_steps):
arm = gaussian_bandit.choose()
reward = 1.0 if random.random() < 0.2 else 0.0
gaussian_bandit.update(Feedback(arm_id=arm.arm_id, reward=reward))
gaussian_bandit.reset()
assert all([arm.squared_reward_sum == 2 and arm.reward_sum == 2 and arm.n == 2 for arm in
gaussian_bandit.arms_dict.values()])
def test_should_run_for_several_iterations_and_sample_all_arms(self):
arm_names = ['arm1', 'arm2', 'arm3']
arms = [GaussianArm(name, n=5, reward_sum=5, squared_reward_sum=5) for name in arm_names]
gaussian_bandit = GaussianThompsonBandit('my_bandit', arms)
num_steps = 1000
sampled_arm_ids = set()
for _ in range(num_steps):
arm = gaussian_bandit.choose()
sampled_arm_ids.add(arm.arm_id)
reward = 1.0 if random.random() < 0.2 else 0.0
gaussian_bandit.update(Feedback(arm.arm_id, reward))
assert sampled_arm_ids == set(arm_names)
def test_should_run_for_several_iterations_and_sample_all_arms(self):
arm_names = ['arm1', 'arm2', 'arm3']
arms = [BetaArm(name, n=2, n_rewards=1) for name in arm_names]
beta_bandit = BetaThompsonBandit('my_bandit', arms)
num_steps = 1000
sampled_arm_ids = set()
for _ in range(num_steps):
arm = beta_bandit.choose()
sampled_arm_ids.add(arm.arm_id)
reward = 1.0 if random.random() < 0.2 else 0.0
beta_bandit.update(Feedback(arm.arm_id, reward))
assert sampled_arm_ids == set(arm_names)
def test_should_reset_value_to_default(self):
arm_names = ['arm1', 'arm2', 'arm3']
arm = [EpsilonGreedyArm(name, n=1, reward_sum=1) for name in arm_names]
egreedy = EGreedy('my_bandit', arm, epsilon=0.1)
num_steps = 20
for _ in range(num_steps):
arm = egreedy.choose()
reward = 1.0 if random.random() < 0.2 else 0.0
egreedy.update(Feedback(arm_id=arm.arm_id, reward=reward))
egreedy.reset()
assert all([
arm.reward_sum == 0 and arm.n == 0.001
for arm in egreedy.arms_dict.values()
])