# Importing the libraries 导入库
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
# 使图像能够调整
%matplotlib notebook 
#中文字体显示  
plt.rc('font', family='SimHei', size=8)

dataset = pd.read_csv('Ads_CTR_Optimisation.csv') # 数据表示虚拟环境，模拟我将投放哪些广告
dataset

import random
N = 10000  # 1000个用户
d = 10     # 10个广告
ads_selected = [] # 广告选择
total_reward = 0 
for n in range(0, N):      # 每个用户循环
    ad = random.randrange(d) # 随机选择广告
    ads_selected.append(ad)  # 将选择的广告加入list中
    reward = dataset.values[n, ad] # 取出数据集中n行ad列查看是否命中，命中值为1，未命中值为0（1即为奖励）
    total_reward = total_reward + reward  # 每轮奖励累计相加

print(total_reward)
# 画图
plt.hist(ads_selected)
plt.title(u'广告选择直方图')
plt.xlabel(u'广告')
plt.ylabel(u'每个广告的点击数')
plt.show()

import math
N = 10000  # 1000个用户
d = 10     # 10个广告
ads_selected = [] # 广告选择
numbers_of_selections = [0] * d # 多项选择
sums_of_rewards = [0] * d  # 奖励总和
total_reward = 0
for n in range(0, N): # 第n个用户
    ad = 0 # 广告初始化
    max_upper_bound = 0  # 最大上界初始化
    for i in range(0, d): # 第i个广告
        if (numbers_of_selections[i] > 0):
            average_reward = float(sums_of_rewards[i]) / float(numbers_of_selections[i]) # 平均奖励，这里如果python2记得用float
            delta_i = math.sqrt(3/2 * math.log(n + 1) / numbers_of_selections[i]) # 置信区间
            upper_bound = average_reward + delta_i # 置信区间上界
#             print(average_reward)
#             print(delta_i )
        else:
            upper_bound = 10000
        if upper_bound > max_upper_bound:
            max_upper_bound = upper_bound
            ad = i
#     print(ad)
    ads_selected.append(ad)
    numbers_of_selections[ad] = numbers_of_selections[ad] + 1
    reward = dataset.values[n, ad]
    sums_of_rewards[ad] = sums_of_rewards[ad] + reward
    total_reward = total_reward + reward
# print(ads_selected)
print(total_reward)
# print(numbers_of_selections)
# print(sums_of_rewards)

# 画图
plt.hist(ads_selected)
plt.title(u'广告选择直方图')
plt.xlabel(u'广告')
plt.ylabel(u'每个广告的点击数')
plt.show()