def main(args, log_dir):
learner = network(args.num_layers, args.num_hidden, args.num_bandits)
train_envs = []
test_envs = []
for i in range(args.num_tasks_train):
train_envs.append(bandit(args, i))
for i in range(args.num_tasks_test):
good_num = np.random.randint(0, args.num_bandits)
test_envs.append(bandit(args, good_num))
#learn(learner,args,train_envs,test_envs,log_dir)
test(learner, args, train_envs, test_envs, log_dir)
#save(args)
return
def generate_graph(todo):
start = time.time()
#jeeeezzz
inst = 0
for i in instances:
print(i)
for algo in todo:
#f = open("../data-for-graph/"+"i-"+i[-5]+"/"+str(algo)+".txt","a")
#g = open("../data-for-graph/"+str(algo)+".txt","a")
#g.write(i+"\n")
print(algo)
for hz in horizons:
regret = 0.0
for seed in range(50):
args = [i, algo, seed, epsilon, hz]
bandit_instance = bandit(args)
REG = bandit_instance.run()
regret += REG
#write file
#f.write(i+' '+algo+' '+str(hz)+' '+str(seed)+' '+str(REG)+'\n')
#print progress
sys.stdout.write("\rseed: %i, time elapsed %.2f" %
(seed, (time.time() - start)))
sys.stdout.flush()
regret /= 50.0
print("\nhorizon:", hz, "Regret:", regret)
#g.write(i+" ----"+str(regret)+"\n")
#f.close()
#g.close()
print("time taken:", time.time() - start)
def upperConfidenceBound(ucb):
averageRewards = np.zeros(ts)
for num in range(run):
bnd = bandit(variance=variance, min=limMin, max=limMax, ucb=ucb)
for t in range(ts):
averageRewards[t] += bnd.takeAction()
return averageRewards / run
def optimisticInitialValues(initial):
averageRewards = np.zeros(ts)
for num in range(run):
bnd = bandit(variance=variance, min=limMin, max=limMax, initial=initial)
for t in range(ts):
averageRewards[t] += bnd.takeAction()
return averageRewards / run
def epsilonGreedyPolicy(epsilon):
averageRewards = np.zeros(ts)
for num in range(run):
bnd = bandit(variance=variance, min=limMin, max=limMax, epsilon=epsilon)
for t in range(ts):
averageRewards[t] += bnd.takeAction()
return averageRewards / run
def generate_graph(todo):
start = time.time()
inst = 0
for i in instances:
print(i)
x = {}
y = {}
for algo in todo:
x[algo] = []
y[algo] = []
f = open("outputDataT2.txt", "a")
#f = open("../data-for-graph/"+"i-"+i[-5]+"/"+str(algo)+".txt","a")
#g = open("../data-for-graph/"+str(algo)+".txt","a")
#g.write(i+"\n")
#print(algo)
for hz in horizons:
regret = 0.0
for seed in range(50):
args = [i, algo, seed, epsilon, hz]
bandit_instance = bandit(args)
REG = bandit_instance.run()
regret += REG
#write file
f.write(i + ', ' + algo + ', ' + str(seed) + ', ' +
str(epsilon) + ', ' + str(hz) + ', ' + str(REG) +
'\n')
#print progress
# sys.stdout.write("\rseed: %i, time elapsed %.2f" % (seed ,(time.time()-start)))
# sys.stdout.flush()
regret /= 50.0
x[algo].append(math.log(hz))
y[algo].append(regret)
#print("\nhorizon:", hz, "Regret:", regret)
#g.write(i+" ----"+str(regret)+"\n")
f.close()
plt.title("Task 2 Comparision for: " + i)
plt.plot(x[todo[0]], y[todo[0]], label='thompson-sampling')
plt.plot(x[todo[1]], y[todo[1]], label='thompson-sampling-with-hint')
plt.legend(loc='upper left', frameon=True)
plt.ylabel('Regret')
plt.xlabel('Horizon (log scale)')
plt.show()
#g.close()
print("time taken:", time.time() - start)
def generate_graph(todo):
start = time.time()
#jeeeezzz
inst = 0
for i in instances:
print(i)
print(todo)
for epsilon in epsilons:
regret = 0.0
for seed in range(50):
args = [i, todo, seed, epsilon, horizon]
bandit_instance = bandit(args)
REG = bandit_instance.run()
regret += REG
sys.stdout.write("\rseed: %i, time elapsed %.2f" %
(seed, (time.time() - start)))
sys.stdout.flush()
regret /= 50.0
print("\nepsilon:", epsilon, "Regret:", regret)
print("time taken:", time.time() - start)
#asks user for number of enemies
numEnemies = int(input('How many monsters will %s battle?\n' % Hero.getName()))
#list that contains enemies
enemyList = []
#list that contains random numbers that are associated with enemies
enemyNumList = []
#sets a random seed with value 0
random.seed(0)
#randomly puts enemies in enemyList, but does so with regards to varying enemy strengths
for i in range(numEnemies):
enemyNumList.append(random.randint(1, 14))
for i in range(numEnemies):
Y = enemyNumList[i]
if Y <= 5:
enemyList.append(bandit())
elif Y <= 9:
enemyList.append(soldier76())
elif Y <= 12:
enemyList.append(zergling())
else:
enemyList.append(illidan())
#sets maximum health for hero as starting health
startHealth = Hero.getHealth()
#hero and enemy combat, one by one
for enemy in enemyList:
print('\nYou have encountered a %s!' % enemy)
enemyHealth = enemy.getHealth()
while True:
import numpy as np
import bandit as b
import matplotlib.pyplot as plt
#constants
numArms = 10
excess = 0.2
neuronsPerArm = 1
epsilon = 0.08
tEpoch = 128
epochs = 2000
#setup and run bandit
p_reward = b.pick_weights(numArms, excess)
bestarm = np.argmax(p_reward)
btest = b.bandit(numArms, neuronsPerArm, tEpoch, epochs, probabilities=p_reward, epsilon=epsilon, recordWeights=False)
(choices, rewards, spikes) = btest.run(epochs)
btest.stop()
print("Plotting results...")
#plot
fig = plt.figure()
plt.scatter(np.arange(epochs), choices, alpha=0.1)
plt.ylabel("Arm Chosen")
plt.xlabel("Time step")
fig.savefig("choices_over_time.png")
fig = plt.figure()
plt.plot(p_reward/100)
plt.hist(choices, bins=np.arange(numArms)-0.5, density=True)
from bandit import bandit
from epsi_greedyPolicy import epsi_greedyPolicy
from ucbPolicy import ucbPolicy
if __name__ == "__main__":
k = 15 # Number of bandits
NUM_CYCLES = 500 # Number of cycles
# Initialize bandits
bandits = []
means = np.random.randint(-3, high=3, size=k)
best_bandit = [0]
best_bandit_val = means[0]
for i in range(k):
bandits.append(bandit(means[i], 1))
if means[i] > best_bandit_val:
best_bandit = [i]
best_bandit_val = means[i]
elif means[i] == best_bandit_val:
best_bandit.append(i)
INITIAL_VAL = 0
policies = [
epsi_greedyPolicy(0.5, range(k), INITIAL_VAL),
epsi_greedyPolicy(0.1, range(k), INITIAL_VAL),
epsi_greedyPolicy(0.01, range(k), INITIAL_VAL),
epsi_greedyPolicy(0.0, range(k), INITIAL_VAL),
ucbPolicy(1, range(k), INITIAL_VAL)
]
num_policies = len(policies)