Pandas多特征对应多标签处理：基于众数生成唯一标签映射

Hey there! No worries at all—every Pandas user starts out asking these kinds of questions, so yours is totally valid and totally worth answering 😊

Here's a straightforward way to create that unique label mapping based on the mode of label for each feature2 + feature3 combination:

Step 1: Set up your data (if you haven't already) #

First, let's get your sample data into a Pandas DataFrame:

import pandas as pd

# Your sample data
data = {
    'feature1': ['a', 'b', 'c'],
    'feature2': [2, 2, 2],
    'feature3': [3, 3, 3],
    'label': [1, 1, 0]
}
df = pd.DataFrame(data)

Step 2: Calculate mode per feature combination #

We'll group the data by feature2 and feature3, then compute the mode of the label column for each group. This gives us the most frequent label for each combination:

# Calculate mode for each (feature2, feature3) pair
# Using lambda to handle cases where multiple modes exist (we pick the first one)
label_mapping = df.groupby(['feature2', 'feature3'])['label'].agg(lambda x: x.mode().iloc[0]).reset_index()

Step 3: Convert to a lookup dictionary (optional but handy) #

If you want a quick way to look up the label for any feature2 + feature3 pair, convert the mapping to a dictionary where the key is a tuple of the two features:

mapping_dict = label_mapping.set_index(['feature2', 'feature3'])['label'].to_dict()

For your sample data, this dictionary will look like:

{(2, 3): 1}

Step 4: Apply the mapping to your data (if needed) #

If you want to add this mapped label back to your original DataFrame, you can do this with apply:

df['mapped_label'] = df.apply(lambda row: mapping_dict[(row['feature2'], row['feature3'])], axis=1)

Your updated DataFrame will now have a mapped_label column with the mode-based label for each row:

Notes on edge cases #

• If a feature2 + feature3 combination has multiple modes (e.g., two labels that appear equally often), the lambda x: x.mode().iloc[0] will pick the first one in the sorted order. If you need a different behavior (like picking the smallest label, or handling it explicitly), you can adjust the lambda function accordingly.
• Since you mentioned feature1 is a unique ID with no actual use, you can drop it anytime with df = df.drop('feature1', axis=1) if it's cluttering your data.

Hope this helps you out! Feel free to follow up if you run into any snags with more complex data or edge cases.

内容的提问来源于stack exchange，提问作者Kevin