如何按组大小将信用卡交易分组数据划分为四分位数并导出CSV

Got it, let's walk through how to split your grouped credit card transactions into quartiles by group size and save each group to the correct directory structure. Here's a step-by-step solution that builds on your existing code:

First, we need to assign each group to a quartile based on its size. We'll use pd.qcut to split groups into 4 equal-sized buckets (or use pd.cut if you prefer fixed value ranges instead of equal counts).

import pandas as pd
import os

# Your existing code to load and group data
# raw_data = pd.read_csv(dataset_path)
raw_data = pd.read_csv('https://query.data.world/s/xapduodvxrzul6tjrupmlqiot4bpex')
categories = ('Cardholder Last Name', 'Cardholder First Initial')
groups = raw_data.groupby(list(categories))
size_of_groups = groups.size()

# Assign quartile labels to each group
quartile_labels = ['1st_quartile', '2nd_quartile', '3rd_quartile', '4th_quartile']
# qcut splits groups into 4 equal-count buckets
group_quartiles = pd.qcut(size_of_groups, q=4, labels=quartile_labels)

# Combine group sizes and quartile labels for easy lookup
group_metadata = pd.DataFrame({'group_size': size_of_groups, 'quartile': group_quartiles})

We'll use Python's os module to automatically create the required folder hierarchy—no manual folder setup needed:

root_directory = "Groups"

# Create each quartile subdirectory (ignore if already exists)
for quartile in quartile_labels:
    target_dir = os.path.join(root_directory, quartile)
    os.makedirs(target_dir, exist_ok=True)

Loop through each grouped dataset, look up its assigned quartile, and save it to the corresponding folder with the requested filename format:

for group_key, group_data in groups:
    # Get the quartile for the current group
    assigned_quartile = group_metadata.loc[group_key, 'quartile']
    # Generate filename from the group key (e.g., ('Mason', 'C') becomes "(Mason, C).csv")
    filename = f"{group_key}.csv"
    # Build full save path
    save_path = os.path.join(root_directory, assigned_quartile, filename)
    # Save the group data to CSV (exclude index for cleaner output)
    group_data.to_csv(save_path, index=False)

If you want to split groups by fixed size ranges instead of equal counts (e.g., groups with size ≤ Q1 go to 1st quartile), use pd.cut instead of pd.qcut:

# Calculate fixed quartile thresholds
q1 = size_of_groups.quantile(0.25)
q2 = size_of_groups.quantile(0.5)
q3 = size_of_groups.quantile(0.75)

# Assign quartiles based on value ranges
group_quartiles = pd.cut(
    size_of_groups,
    bins=[-float('inf'), q1, q2, q3, float('inf')],
    labels=quartile_labels,
    include_lowest=True
)

To check if groups are distributed as expected, run this to see the count of groups per quartile:

print(group_metadata['quartile'].value_counts())

内容的提问来源于stack exchange，提问作者user13051721