Python：将多列二进制数据转换为单列分类列

Alright, let's tackle this CSV compression problem step by step. You've got a 170-column dataset where 5 are unique identifiers, and the rest are binary features split across 10 categories—your goal is to shrink this down to 15 columns total (5 identifiers + 10 category columns). Here's how to do it cleanly with pandas:

核心思路 #

The key here is to group all binary features by their 10 categories, then aggregate each group into a single column. Since you're working with binary data, there are a few common aggregation approaches depending on what you need from the compressed data.

Step 1: Load & Prep Your Data #

First, load your CSV and separate the identifier columns from the binary feature columns:

import pandas as pd

# Load your actual CSV file
df = pd.read_csv("your_dataset.csv")

# Define your 5 unique identifier columns (match your actual column names exactly!)
id_columns = ["Platform", "ID", "date", "通话时长", "name"]

# Extract all binary feature columns (everything not in the identifier list)
feature_columns = [col for col in df.columns if col not in id_columns]

Step 2: Group Features by Category #

Next, you need to map each binary feature to its corresponding category. This depends on how your feature columns are named—most datasets use consistent prefixes (e.g., Billing_issue1, Billing_issue2 for a "Billing" category). Here's how to automate this if your columns follow a pattern:

# Example: If your 10 categories are named like "Category1", "Category2", ..., "Category10"
# And feature columns start with the category name (e.g., "Category1_featureA")
category_groups = {}
for category in ["Category1", "Category2", "Category3", "Category4", "Category5", 
                 "Category6", "Category7", "Category8", "Category9", "Category10"]:
    # Grab all features that start with this category's name
    category_groups[category] = [col for col in feature_columns if col.startswith(category)]

# If your columns don't have a consistent prefix, you'll need to manually define the groups:
# category_groups = {
#     "Billing": ["late_fee", "payment_failed", "invoice_error"],
#     "Technical": ["app_crash", "login_failure", "slow_load"],
#     # ... add the remaining 8 categories
# }

Step 3: Aggregate Each Category into One Column #

Now, compress each category's features into a single column. Choose the aggregation method that fits your use case:

Option 1: Count of Active Features (Most Common) #

Count how many binary features are set to 1 for each category:

# Start with just the identifier columns
compressed_df = df[id_columns].copy()

# Add aggregated columns for each category
for category_name, feature_list in category_groups.items():
    compressed_df[category_name] = df[feature_list].sum(axis=1)

Option 2: Binary Flag (Any Active Feature) #

If you only care whether at least one feature in the category is 1 (instead of the count), use this:

for category_name, feature_list in category_groups.items():
    compressed_df[category_name] = df[feature_list].any(axis=1).astype(int)

Option 3: List of Active Features #

If you need to track which features were active (not just count/flag), concatenate the feature names:

for category_name, feature_list in category_groups.items():
    # For each row, join the names of features that are 1
    compressed_df[category_name] = df[feature_list].apply(
        lambda row: ", ".join([col for col, val in row.items() if val == 1]),
        axis=1
    )

Step 4: Verify the Result #

Check that your compressed dataframe has exactly 15 columns:

print(f"Compressed columns count: {len(compressed_df.columns)}")  # Should output 15

Quick Notes #

• Double-check that your category_groups correctly maps all 165 feature columns to the 10 categories (no missing or misassigned columns!).
• If your binary data uses values other than 0/1 (e.g., True/False), adjust the aggregation logic accordingly (e.g., sum() still works for booleans in pandas).

内容的提问来源于stack exchange，提问作者REFER