Python特征选择：如何用内置方法查找选中特征在全特征中的列索引

Hey there! When you're knee-deep in feature selection and need to map your selected features back to their original column indices in the full feature set, Python doesn't have a single "one-size-fits-all" built-in function—but combining basic built-in tools or using standard libraries makes this super straightforward. Let's break down the best approaches based on your use case:

1. Pure Python (No External Libraries) #

If you're working with standard lists (your "arrays" are Python lists), the index() method combined with a list comprehension is the simplest way to get matching indices.

Basic Implementation #

# Example full feature list (large array)
full_features = ['age', 'income', 'gender', 'education', 'occupation']
# Selected features (small array)
selected_features = ['gender', 'age', 'occupation']

# Get indices in the order of selected_features
indices = [full_features.index(feat) for feat in selected_features]
print(indices)  # Output: [2, 0, 4]

Handle Missing Features Gracefully #

If there's a chance some selected features aren't in the full list (easy to happen with typos!), add error handling to avoid crashes:

indices = []
for feat in selected_features:
    try:
        idx = full_features.index(feat)
        indices.append(idx)
    except ValueError:
        print(f"⚠️ Warning: Feature '{feat}' not found in the full feature list")

2. NumPy for Large/Array Data #

If you're working with NumPy arrays (common in data science workflows), vectorized operations will be much faster for large datasets.

Basic Matching (Preserve Selected Order) #

import numpy as np

full_features = np.array(['age', 'income', 'gender', 'education', 'occupation'])
selected_features = np.array(['gender', 'age', 'occupation'])

# Get indices matching the order of selected_features
indices = np.argsort(full_features)[np.searchsorted(full_features, selected_features)]
print(indices)  # Output: [2 0 4]

Find All Matching Indices (For Duplicate Features) #

If your full feature list has duplicate names (rare but possible), use np.isin to grab all matching positions:

full_features = np.array(['age', 'income', 'age', 'gender', 'income'])
selected_features = np.array(['age', 'income'])

# Get all indices where features match
all_indices = np.where(np.isin(full_features, selected_features))[0]
print(all_indices)  # Output: [0 1 2 4]

3. Handling Duplicate Features (Pure Python) #

If you need to track every occurrence of a selected feature in the full list (not just the first one), use a dictionary to collect indices:

full_features = ['age', 'income', 'age', 'gender', 'income']
selected_features = ['age', 'income']

feature_indices = {feat: [] for feat in selected_features}
for idx, feat in enumerate(full_features):
    if feat in feature_indices:
        feature_indices[feat].append(idx)

print(feature_indices)  # Output: {'age': [0, 2], 'income': [1, 4]}

Key Notes #

• Use the pure Python approach for small datasets or when you want minimal dependencies.
• Opt for NumPy if you're working with large arrays—it's optimized for speed.
• Always account for missing or duplicate features if your data might have them!

内容的提问来源于stack exchange，提问作者Khan