def getReward(self, arm, reward):
UCB.getReward(self, arm, reward)
self.budget += reward
self.estmeans[arm] = (self.estmeans[arm] *
(self.pulls[arm] - 1) + reward) / self.pulls[arm]
if (reward > 0):
self.successes[arm] += 1
def __init__(self,
nbArms,
inibudget=10.0,
safebudget=1.0,
lower=-1.0,
amplitude=2.0):
UCB.__init__(self, nbArms, lower=lower, amplitude=amplitude)
SafeAlg.__init__(self, nbArms)
def __init__(self,
nbArms,
inibudget=10.0,
min_r=-1.0,
max_r=+1.0,
lower=0.0,
amplitude=1.0):
UCB.__init__(self, nbArms, lower=lower, amplitude=amplitude)
Budgeted.__init__(self, inibudget=inibudget, min_r=min_r, max_r=max_r)
BernoulliEstimator.__init__(self, nbArms)
def getReward(self, arm, reward):
UCB.getReward(self, arm, reward)
if (reward >= 0):
self.positive_count[arm] += 1
self.positive_mean[arm] = (self.positive_mean[arm] *
(self.positive_count[arm] - 1) +
reward) / self.positive_count[arm]
else:
self.negative_count[arm] += 1
self.negative_mean[arm] = (self.negative_mean[arm] *
(self.negative_count[arm] - 1) +
reward) / self.negative_count[arm]
def __init__(self,
nbArms,
inibudget=10.0,
safebudget=1.0,
min_r=-1.0,
max_r=+1.0,
lower=0.0,
amplitude=1.0):
UCB.__init__(self, nbArms, lower=lower, amplitude=amplitude)
SafeAlg.__init__(self,
nbArms,
inibudget=inibudget,
min_r=min_r,
max_r=max_r,
safebudget=safebudget)
""" Example of use of SMPyBandits.
See https://SMPyBandits.GitHub.io/API.html for more details!"""
import numpy as np
np.random.seed(0) # for reproducibility
from SMPyBandits.Arms import Bernoulli
arms = [Bernoulli(0.1), Bernoulli(0.9)]
from SMPyBandits.Environment import MAB
my_MAB_problem = MAB(arms)
nbArms = my_MAB_problem.nbArms # 2 arms !
from SMPyBandits.Policies import UCB
my_UCB_algo = UCB(nbArms)
my_UCB_algo.startGame() # reset internal memory
horizon = 1000
for t in range(horizon): # simulation loop
chosen_arm = my_UCB_algo.choice()
observed_reward = my_MAB_problem.draw(chosen_arm)
my_UCB_algo.getReward(chosen_arm, observed_reward)
cumulated_reward = sum(my_UCB_algo.rewards) # random!
number_of_plays = sum(my_UCB_algo.pulls) # horizon = 1000
mean_reward = cumulated_reward / number_of_plays
print("The UCB algorithm obtains here a mean reward =", mean_reward)
def startGame(self):
UCB.startGame(self)
self.budget = self.inibudget
self.estmeans.fill(0.0)
self.successes.fill(0)
def startGame(self):
UCB.startGame(self)
Budgeted.startGame(self)
BernoulliEstimator.startGame(self)
def getReward(self, arm, reward):
UCB.getReward(self, arm, reward)
self.budget += reward
def __init__(self, nbArms, inibudget=10.0, lower=-1.0, amplitude=2.0):
UCB.__init__(self, nbArms, lower=lower, amplitude=amplitude)
self.inibudget = inibudget
self.budget = inibudget
self.estmeans = np.zeros(nbArms)
self.successes = np.zeros(nbArms, dtype='int')
def getReward(self, arm, reward):
UCB.getReward(self, arm, reward)
SafeAlg.getReward(self, arm, reward)
def choice(self):
r = SafeAlg.choice(self)
if r is None:
r = UCB.choice(self)
return r
def startGame(self):
UCB.startGame(self)
SafeAlg.startGame(self)
def getReward(self, arm, reward):
UCB.getReward(self, arm, reward)
Budgeted.getReward(self, reward)
BernoulliEstimator.getReward(self, arm, reward)
def startGame(self):
UCB.startGame(self)
self.reward_samples = [np.array([0.0]) for a in range(self.nbArms)]
def startGame(self):
UCB.startGame(self)
self.positive_mean = nb.repeat(0.0, self.nbArms, dtype='float')
self.negative_mean = nb.repeat(0.0, self.nbArms, dtype='float')
self.positive_count = nb.repeat(0, self.nbArms, dtype='int')
self.negative_count = nb.repeat(0, self.nbArms, dtype='int')
def getReward(self, arm, reward):
UCB.getReward(self, arm, reward)
self.reward_samples[arm] = np.sort(
np.append(self.reward_samples[arm], [reward]))