根据列值拆分 Python 数据框，然后在算法中使用它们

Question

我目前正在使用来自 mlxtend 的 Apriori 算法进行简单的频繁模式分析。目前，我只是在查看所有交易。但我想根据国家区分分析。我当前的脚本如下所示：

import pandas as pd
import numpy as np
import pyodbc
from mlxtend.preprocessing import TransactionEncoder
from mlxtend.frequent_patterns import apriori
from mlxtend.frequent_patterns import association_rules

dataset = pd.read_sql_query("""some query"", cnxn)

# Transform/prep dataset into list data
dataset_tx = dataset.groupby(['ReceiptCode'])['ItemCategoryName'].apply(list).values.tolist()

# Define classifier
te = TransactionEncoder()

# Binary-transform dataset
te_ary = te.fit(dataset_tx).transform(dataset_tx)

# Fit to new dataframe (sparse dataframe)
df = pd.SparseDataFrame(te_ary, columns=te.columns_)

# Run algorithm 
frequent_itemsets = apriori(df, min_support=0.10, use_colnames=True)
frequent_itemsets['length'] = frequent_itemsets['itemsets'].apply(lambda x: len(x))
rules = association_rules(frequent_itemsets, metric="confidence", min_threshold=0.3)

下面是dataset的例子。

+----------------------+--+------------------+--+------------------+
|     ReceiptCode      |  | ItemCategoryName |  | StoreCountryName |
+----------------------+--+------------------+--+------------------+
|  0000P70322000031467 |  |  Food            |  |   Denmark        |
|  0000P70322000031867 |  |  Food            |  |   Denmark        |
|  0000P70322000051467 |  |  Interior        |  |   Germany        |
|  0000P70322000087468 |  |  Kitchen         |  |   Switzerland    |
|  0000P70322000031469 |  |  Leisure         |  |   Germany        |
|  0000P70322000031439 |  |  Food            |  |   Switzerland    |
+----------------------+--+------------------+--+------------------+

是否可以"automatically"基于列StoreCountryName创建多个数据框，然后在算法中使用它，即在分析中使用特定国家/地区的数据框并遍历所有国家/地区？我知道我可以手动创建数据框，然后应用转换和分析。

Answer 1

您可以 groupby 并进行列表理解以将数据帧存储在列表中，然后遍历它们：

g = df.groupby('StoreCountryName')
dfs = [group for _,group in g]

for i in range(len(dfs)):
    dfs[i]['iteration'] = i # do stuff to each frame
    print(f"{dfs[i]} \n")

           ReceiptCode ItemCategoryName StoreCountryName  iteration
0  0000P70322000031467             Food          Denmark          0
1  0000P70322000031867             Food          Denmark          0 

           ReceiptCode ItemCategoryName StoreCountryName  iteration
2  0000P70322000051467         Interior          Germany          1
4  0000P70322000031469          Leisure          Germany          1 

           ReceiptCode ItemCategoryName StoreCountryName  iteration
3  0000P70322000087468          Kitchen      Switzerland          2
5  0000P70322000031439             Food      Switzerland          2

或者您可以创建一个函数并使用 groupby 和 apply

def myFunc(country):
    # do stuff

df.groupby('StoreCountryName').apply(myFunc)

根据列值拆分 Python 数据框，然后在算法中使用它们

Splitting up a Python dataframe based on column value and then using them in algorithm

python

apriori

pandas