thompson_sampling_snippet.py

from scipy.stats import beta
import random

ads_selected = []
# d = number of ads
number_of_wins = [0] * d
number_of_losses = [0] * d

total_reward = 0

# For each user
for n in range(0, N):
    ad = 0
    max_random = 0
    # Iterate over ads
    for i in range(0, d):
        # Randomly pull a value from the betavariate distribution for that ad
        random_beta = random.betavariate(number_of_wins[i] + 1, number_of_losses[i] + 1)
        # Selecting the highest value
        if random_beta > max_random:
            max_random = random_beta
            ad = i
    # Selecting the winning ad for the current user
    ads_selected.append(ad)
    
    # If the user has clicked, update the number of successes for that ad
    # Else update number of failures
    reward = df.values[n, ad]
    if reward == 1:
        number_of_wins[ad] = number_of_wins[ad] + 1
    else:
        number_of_losses[ad] = number_of_losses[ad] + 1
    total_reward = total_reward + reward