我正在使用 Python product recommendation system (参见答案 Mohsin hasan)。
简单的脚本将两个变量(UserId、ItemId)作为输入,并将两个产品之间的亲和性分数作为输出。
但是,我添加了第三列(国家/地区)。 我想针对每个国家(而不是整个数据框)单独进行分析。
最初,我使用 R,其中 dplyr 的“group_by”函数应该有所帮助。但是目前我被卡住了(请参阅下面的尝试)。任何人都知道我如何对每个国家/地区进行此分析? (我觉得 'pandas.DataFrame.groupby' 也可以解决这个问题,而不是我尝试使用 for 循环)。
示例数据(请注意:唯一的区别是我添加了国家列:
UserId ItemId Country
1 Babyphone Netherlands
1 Babyphone Netherlands
1 CoffeeMachine Netherlands
2 CoffeeMachine Netherlands
2 Shaver Netherlands
3 Shaver Netherlands
3 CoffeeMachine Netherlands
4 CoffeeMachine Netherlands
4 Shaver Netherlands
4 Blender Netherlands
5 Blender Netherlands
5 BabyPhone Netherlands
5 Shaver Netherlands
6 Shaver Netherlands
7 CoffeeMachine Netherlands
7 CoffeeMachine Netherlands
8 BabyPhone Netherlands
9 Blender Netherlands
9 Blender Netherlands
1 Babyphone Germany
1 Babyphone Germany
1 CoffeeMachine Germany
2 CoffeeMachine Germany
2 Shaver Germany
3 Shaver Germany
3 CoffeeMachine Germany
4 CoffeeMachine Germany
4 Shaver Germany
4 Blender Germany
5 Blender Germany
5 BabyPhone Germany
5 Shaver Germany
6 Shaver Germany
7 CoffeeMachine Germany
7 CoffeeMachine Germany
8 BabyPhone Germany
9 Blender Germany
9 Blender Germany
工作 - 原始 - 代码(使用 UserId 和 ItemId,没有国家/地区)
# main is our data.
# get unique items
items = set(main.productId)
n_users = len(set(main.userId))
# make a dictionary of item and users who bought that item
item_users = main.groupby('productId')['userId'].apply(set).to_dict()
# iterate over combinations of item1 and item2 and store scores
result = []
for item1, item2 in itertools.combinations(items, 2):
score = len(item_users[item1] & item_users[item2]) / n_users
item_tuples = [(item1, item2), (item2, item1)]
result.append((item1, item2, score))
result.append((item2, item1, score)) # store score for reverse order as well
# convert results to a dataframe
result = pd.DataFrame(result, columns=["item1", "item2", "score"])
我的尝试(在国家/地区,但它不起作用)。我尝试了什么?
- 过滤每个国家的数据框(是的,这很糟糕,因为它不是动态的)
- 遍历数据框(每个国家/地区都有 1 个数据框)
- 尝试插入解决方案(见上文),并单独申请数据框。
如您所见,很遗憾,它无法正常工作...
Netherlands = df.loc[df['Country'] == 'Netherlands'] Germany = df.loc[df['Country'] == 'Germany'] results = [] for dataset in (Netherlands, Germany): for index, row in dataset.iterrows(): Country = row['Country'] # Need to bind the name of the df later to the results items = set(dataset.ItemId) #Get unique Items per country n_users = len(set(dataset.UserId) # Get unique number of users per country item_users = dataset.groupby('ItemId'['UserId'].apply(set).to_dict() # I tried to add country here, but without results. for item1, item2 in itertools.combinations(items, 2): print("item1", item1) print("item2", item2) score = len(item_users[item1] & item_users[item2]) / n_users item_tuples = [(item1, item2), (item2, item1)] result.append((item1, item2, score)) result.append((item2, item1, score)) # store score for reverse order as well result = pd.DataFrame(result, columns=["item1", "item2", "score"])
Edit1:预期输出
编辑2 分数是如何计算的? 分数代表:有多少客户一起购买产品组合。
例如,在数据中您看到 Razor 和咖啡机 = 0.333(因为每个国家/地区 9 个人中有 3 人购买了此组合)。在第一个代码中,乐谱运行良好。但是,我无法在每个国家/地区运行它(这是这里的关键问题)。
非常感谢!
请您参考如下方法:
给你
=^..^=
正如您提到的那样,group by will by used。首先将得分循环移动到带有附加字段“国家/地区”的函数中,然后将其用于分组数据帧,如下所示:
import pandas as pd
import itertools
将分数移动到函数中:
def get_score(item):
country = item[0]
df = item[1]
# get unique items
items = set(df.ItemId)
n_users = len(set(df.UserId))
# make a dictionary of item and users who bought that item
item_users = df.groupby('ItemId')['UserId'].apply(set).to_dict()
# iterate over combinations of item1 and item2 and store scores
result = []
for item1, item2 in itertools.combinations(items, 2):
score = len(item_users[item1] & item_users[item2]) / n_users
item_tuples = [(item1, item2), (item2, item1)]
result.append((item1, item2, score, country))
result.append((item2, item1, score, country)) # store score for reverse order as well
# convert results to a dataframe
result = pd.DataFrame(result, columns=["item1", "item2", "score", 'country'])
return result
按国家/地区对数据进行分组,然后遍历每个组以获得分数:
grouped_data = df.groupby(['Country'])
df_list = []
for item in list(grouped_data):
df_list.append(get_score(item))
# concat frames
df = pd.concat(df_list)
# remove rows with 0 score
df = df[df['score'] > 0]
输出:
item1 item2 score country
0 BabyPhone Blender 0.111111 Germany
1 Blender BabyPhone 0.111111 Germany
4 BabyPhone Shaver 0.111111 Germany
5 Shaver BabyPhone 0.111111 Germany
8 Blender CoffeeMachine 0.111111 Germany
9 CoffeeMachine Blender 0.111111 Germany
10 Blender Shaver 0.222222 Germany
11 Shaver Blender 0.222222 Germany
14 CoffeeMachine Shaver 0.333333 Germany
15 Shaver CoffeeMachine 0.333333 Germany
16 CoffeeMachine Babyphone 0.111111 Germany
17 Babyphone CoffeeMachine 0.111111 Germany
0 BabyPhone Blender 0.111111 Netherlands
1 Blender BabyPhone 0.111111 Netherlands
4 BabyPhone Shaver 0.111111 Netherlands
5 Shaver BabyPhone 0.111111 Netherlands
8 Blender CoffeeMachine 0.111111 Netherlands
9 CoffeeMachine Blender 0.111111 Netherlands
10 Blender Shaver 0.222222 Netherlands
11 Shaver Blender 0.222222 Netherlands
14 CoffeeMachine Shaver 0.333333 Netherlands
15 Shaver CoffeeMachine 0.333333 Netherlands
16 CoffeeMachine Babyphone 0.111111 Netherlands
17 Babyphone CoffeeMachine 0.111111 Netherlands
