def run_experiment(p1, p2, p3, N):
bandits = [Bandit(p1), Bandit(p2), Bandit(p3)]
data = np.empty(N)
for i in range(N):
b_list = [b.sample() for b in bandits]
j = np.argmax(b_list)
x = bandits[j].pull()
bandits[j].update(x)
print(b_list)
data[i] = x
cumulative_average = np.cumsum(data) / (np.arange(N) + 1)
print(cumulative_average)
plt.plot(cumulative_average)
plt.plot(np.ones(N) * p1)
plt.plot(np.ones(N) * p2)
plt.plot(np.ones(N) * p3)
plt.ylim((0, 1))
plt.xscale('log')
plt.legend()
plt.show()
def run_experiment(p1, p2, p3, N): #converge experiment function:
bandits = [
Bandit(p1), Bandit(p2), Bandit(p3)
] #Create 3 bandits with probability p1,p2,p3; we play 100000 times
data = np.empty(N) #store results in data
for i in range(N):
# thompson sampling
j = np.argmax([b.sample() for b in bandits])
x = bandits[j].pull()
bandits[j].update(x)
# for the plot
data[i] = x #x means "Get a reward or not" or "click or not"
cumulative_average_ctr = np.cumsum(data) / (
np.arange(N) + 1
) #we keep track of the cumulative click_through_rate by
#divide the total number of click by the total showing time.
# plot moving average ctr #when we are using Thompson Sampling.
plt.plot(cumulative_average_ctr)
plt.plot(np.ones(N) * p1)
plt.plot(np.ones(N) * p2)
plt.plot(np.ones(N) * p3)
plt.ylim((0, 1))
plt.xscale('log')
plt.show() #What we see here is after 100000 trials, the average ctr
def run_experiment(p1, p2, p3,
N): # 3 probabilities = 3 bandits and number of trials.
# define each bandit
bandits = [Bandit(p1), Bandit(p2), Bandit(p3)]
# keep track of all the data we get. 1 for click, 0 for no-click
data = np.empty(N)
for i in xrange(N):
# thompson sampling
j = np.argmax([b.sample() for b in bandits])
x = bandits[j].pull()
bandits[j].update(x)
# for the plot, keep track of x which is the data at i.
data[i] = x
cumulative_average_ctr = np.cumsum(data) / (np.arange(N) + 1)
# plot moving average ctr
plt.plot(cumulative_average_ctr)
plt.plot(np.ones(N) * p1)
plt.plot(np.ones(N) * p2)
plt.plot(np.ones(N) * p3)
plt.ylim((0, 1))
plt.xscale('log')
plt.show()
def run_experiment(p1, p2, p3, N):
'args: probabilities for 3 bandits and number of experiments.'
bandits = [Bandit(p1), Bandit(p2), Bandit(p3)]
data = np.empty(N)
for i in range(N):
best_bandito = np.argmax([bandito.sample() for bandito in bandits])
result = bandits[best_bandito].pull()
data[i] = result
cumulative_avg_ctr = np.cumsum(data) / (np.arange(N) + 1)
plt.plot(cumulative_avg_ctr)
plt.plot(np.ones(N) * p1, '--')
plt.plot(np.ones(N) * p2, '-.')
plt.plot(np.ones(N) * p3, ':')
plt.ylim((0, .6))
plt.xscale('log')
plt.xlabel('Number of Trials')
plt.show()
def run_experiment(p1, p2, p3, N):
bandits = [Bandit(p1), Bandit(p2), Bandit(p3)]
data = np.empty(N)
for i in range(N):
j = np.argmax([b.sample() for b in bandits]) # Select the best option
x = bandits[j].pull() # Pull this lever
bandits[j].update(x) # Update the distribution
data[i] = x # Store the data
cumulative_average = np.cumsum(data) / (np.arange(N) + 1
) # Find the average
plt.plot(cumulative_average) # Plotting the average of the decision
plt.plot(np.ones(N) * p1) # Plotting of each bandit's true probability
plt.plot(np.ones(N) * p2)
plt.plot(np.ones(N) * p3)
plt.ylim((0, 1))
plt.xscale('log') # Easier to visualize due to # of iterations
plt.show()
def run_experiment(p1, p2, p3, N):
bandits = [Bandit(p1), Bandit(p2), Bandit(p3)]
data = np.empty(N)
for i in range(N):
# thompson sampling
j = np.argmax([b.sample() for b in bandits])
x = bandits[j].pull()
bandits[j].update(x)
# for the plot
data[i] = x
cumulative_average_ctr = np.cumsum(data) / (np.arange(N) + 1)
# plot moving average ctr
plt.plot(cumulative_average_ctr)
plt.plot(np.ones(N)*p1)
plt.plot(np.ones(N)*p2)
plt.plot(np.ones(N)*p3)
plt.ylim((0,1))
plt.xscale('log')
plt.show()
def run_experiment(p1, p2, p3, N): bandits = [Bandit(p1), Bandit(p2), Bandit(p3)] data = np.empty(N)