Epsilon Greedy from "Bandit Algorithms"

EpsilonGreedy.py

import numpy as np
class EpsGreedy()
  def __init__(self, number_of_bandits, epsilon, start_greedy= True):
    self.count = np.zeros(number_of_bandits)
    self.scores = np.array([int(!start_greedy)] * number of bandits)  
    self.epsilon = epsilon  
    self.bandit_count = number_of_bandits
  def select_arm():  
    choice = np.random.binomial(1,self.epsilon):
    if choice = 1: #EXPLORE
      return np.random.randint(self.number_of_bandits)
    else:
      return self.scores.argmax()  
  def update(self, bandit_index, reward):
      self.count[bandit_index] = self.count[bandit_index] + 1
      n = self.count[bandit_index]
      score = self.scores[bandit_index]
      new_value = (n-1/float(n)) * score + (1/float(n)) * reward
      self.score[bandit_index] = new_value
      
class Binomial_Bandit:
  def __init__(self, p_of_payoff):
    self.probability = p_of_payoff
  def get_reward():
    return np.binomial(1,self.probability)