def get_reward_distribution():
global mean_list
mean_list = np.zeros(10)
# get the 10 q*(a)
for i in range(10):
mean_list[i] = rand_norm(0.0, 1.0)
def env_step(this_action): # returns (floating point, NumPy array, Boolean), this_action: NumPy array
global this_reward_observation, rewards
avg_reward = rewards[this_action[0]]
the_reward = rand_norm(avg_reward, 1.0) # rewards drawn from (0, 1) Gaussian
this_reward_observation = (the_reward, this_reward_observation[1], False)
return this_reward_observation
def env_step(
this_action
): # returns (floating point, NumPy array, Boolean), this_action: NumPy array
global this_reward_observation
the_reward = rand_norm(reward_distribution[int(this_action)], 1.0)
this_reward_observation = (the_reward, this_reward_observation[1], False)
return this_reward_observation
def env_step(
this_action
): # returns (floating point, NumPy array, Boolean), this_action: NumPy array
global this_reward_observation
the_reward = rand_norm(distributions[int(this_action)],
1.0) # rewards drawn from (0, 1) Gaussian
this_reward_observation = (the_reward, this_reward_observation[1], False)
return this_reward_observation
def env_init():
global this_reward_observation,q
local_observation = np.zeros(0) # An empty NumPy array
this_reward_observation = (0.0, local_observation, False)
#set up q*(a)
for i in range(0,10):
q[i]=rand_norm(0.0, 1.0)
def env_init():
global this_reward_observation, reward_distribution
local_observation = np.zeros(0) # An empty NumPy array
reward_distribution = np.zeros(10)
for i in range(10):
reward_distribution[i] = rand_norm(0.0, 1.0)
this_reward_observation = (0.0, local_observation, False)
def __init__(self, number_of_arms):
super(KArmedTestbed, self).__init__()
self.number_of_arms = number_of_arms
self.action_value = np.array(
[rand_norm(0, 1.0) for i in np.arange(number_of_arms)])
self.current_state = np.zeros(0)
self.available_actions = np.array(
[i for i in np.arange(1, number_of_arms + 1, 1)])
def createDistribution(
): # create distributions contains mean of each distribution
global distributions, numActions
i = 0
distributions = []
while i < 10:
distribution = rand_norm(0.0, 1.0)
distributions.append(distribution)
i += 1
return
def env_init():
global this_reward_observation, bandits, opt_act
local_observation = np.zeros(0) # An empty NumPy array
bandits = np.zeros(10)
opt_act = 0
for i in xrange(10):
bandits[i] = rand_norm(0, 1)
if bandits[opt_act] < bandits[i]:
opt_act = i
this_reward_observation = (0.0, local_observation, False)
def env_step(
this_action
): # returns (floating point, NumPy array, Boolean), this_action: NumPy array
global mean_q_actions_array
global this_reward_observation
action = int(this_action)
#obtains a reward for the action taken
#the value is based on a distribution of mean = mean value for that action and variance = 1
the_reward = rand_norm(mean_q_actions_array[action], 1)
this_reward_observation = (the_reward, this_reward_observation[1], False)
#returns (reward, state and if some terminal state was hit)
return this_reward_observation
def env_step(this_action): # returns (floating point, NumPy array, Boolean), this_action: NumPy array
global this_reward_observation,real_reward,q
# rewards drawn from (q(At), 1) Gaussian
for i in range(10):
real_reward[i] = rand_norm(q[i], 1)
the_reward = real_reward[this_action]
#print "the reward for action %d is %f, optimal is %d,optimal value is %f"%(int(this_action),the_reward,max(enumerate(real_reward),key=lambda x: x[1])[0],real_reward[max(enumerate(real_reward),key=lambda x: x[1])[0]])
this_reward_observation = (the_reward, this_reward_observation[1], False)
return this_reward_observation
def env_init():
global this_reward_observation
global mean_q_actions_array
global num_actions
#initialize the variables for the enviroment, dependending on the quantity of actions available
num_actions = 10
mean_q_actions_array = np.zeros(num_actions)
#Initialize the mean values for each action utilizing a Gaussian distribution (mean = 0, variance = 1)
for i in range(0, num_actions):
mean_q_actions_array[i] = rand_norm(0.0, 1.0)
local_observation = np.zeros(0) # An empty NumPy array
this_reward_observation = (0.0, local_observation, False)
def env_get_reward(this_action):
global mean_list
the_reward = rand_norm(mean_list[int(this_action[0])], 1.0)
return the_reward
def _arms(self, k):
tmp = np.zeros(k)
for cnt in range(k):
tmp[cnt] = rand_norm(0, 1)
# print('q* is {}'.format(tmp))
return tmp
def q():
global qa
qa = []
for i in range(num_actions):
qa.append(rand_norm(0.0, 1.0))
return
def do_action(self, action):
for a in self.available_actions:
self.action_value[a - 1] += rand_norm(0.0, 0.01)
return rand_norm(self.action_value[action], 1.0)
def do_action(self, action):
# state update goes here
return rand_norm(self.action_value[action], 1.0)
def env_init():
global this_reward_observation
local_observation = np.zeros(0) # An empty NumPy array
for i in range(actions):
qtrue[i] = rand_norm(0.0, 1.0)
this_reward_observation = (0.0, local_observation, False)
def env_init():
global arms_centres
for i in range(arm_count):
arms_centres[i] = rand_norm(0, 1)
def reward(self, action):
# print('Action is ' + str(action))
index = int(action)
return rand_norm(self._q_array[index], 1)
def env_init():
global this_reward_observation, mode, rewards
local_observation = np.zeros(0) # An empty NumPy array
for i in range(10):
rewards[i] = rand_norm(0.0,1.0)
this_reward_observation = (0.0, local_observation, False)
