paulpas Feature Interaction
Comprehensive guide to feature interaction in machine learning and data science workflows.
When to Use This Skill
- Solving real-world feature engineering problems
- Building machine learning pipelines with feature interaction
- Implementing best practices for feature interaction
- Optimizing model performance using feature interaction techniques
- Learning industry-standard approaches to feature interaction
When NOT to Use This Skill
- When using pre-built libraries without understanding underlying concepts
- For toy problems that don't require feature interaction rigor
- When domain expertise in specific problem requires different approach
- If your problem doesn't require the complexity this skill provides
Purpose and Key Concepts
Feature Interaction is a critical component of the machine learning workflow. This skill covers:
- Theoretical foundations — Mathematical principles and statistical concepts
- Practical implementation — Working code examples and patterns
- Common pitfalls — Mistakes to avoid and how to recover from them
- Best practices — Industry-standard approaches and optimization techniques
Core Workflow
- Understand the problem — Clearly define what you're solving for
- Select approach — Choose the right technique for your data and constraints
- Implement solution — Write clean, tested code following best practices
- Validate results — Verify your implementation with tests and validation
- Optimize performance — Improve efficiency and accuracy incrementally
Implementation Patterns
Pattern 1: Basic Feature Interaction
import pandas as pd
import numpy as np
from sklearn.preprocessing import PolynomialFeatures
from typing import List, Tuple
def generate_basic_interactions(df: pd.DataFrame, feature_pairs: List[Tuple[str, str]] = None, degree: int = 2) -> pd.DataFrame:
"""
Generate basic feature interactions including polynomial terms and cross-features.
Args:
df: Input DataFrame containing numerical features
feature_pairs: Optional list of specific column pairs to interact
degree: Degree for polynomial features (default 2)
Returns:
DataFrame with original features plus interaction terms
"""
if df.empty:
raise ValueError("Input DataFrame cannot be empty")
numerical_cols = df.select_dtypes(include=[np.number]).columns.tolist()
if len(numerical_cols) < 2:
raise ValueError("At least two numerical columns are required for interactions")
# Create polynomial features for all numerical columns
poly = PolynomialFeatures(degree=degree, include_bias=False, interaction_only=False)
poly_features = poly.fit_transform(df[numerical_cols])
poly_col_names = poly.get_feature_names_out(numerical_cols)
# Create specific cross-features if provided
cross_features = {}
if feature_pairs:
for col1, col2 in feature_pairs:
if col1 in df.columns and col2 in df.columns:
cross_features[f'{col1}_x_{col2}'] = df[col1] * df[col2]
cross_features[f'{col1}_plus_{col2}'] = df[col1] + df[col2]
# Combine original data with polynomial features
result_df = pd.DataFrame(poly_features, columns=poly_col_names, index=df.index)
# Add cross-features if any
if cross_features:
cross_df = pd.DataFrame(cross_features, index=df.index)
result_df = pd.concat([result_df, cross_df], axis=1)
return result_df
# Self-contained test
if __name__ == "__main__":
test_df = pd.DataFrame({
'A': np.random.randn(50)
'B': np.random.randn(50)
'C': np.random.randn(50)
})
result = generate_basic_interactions(test_df, feature_pairs=[('A', 'B')])
print(f"Generated {result.shape[1]} features from {test_df.shape[1]} original columns")
Pattern 2: Production-Ready Feature Interaction
import logging
import pandas as pd
import numpy as np
from typing import Any, Dict, List
from sklearn.preprocessing import PolynomialFeatures
logger = logging.getLogger(__name__)
class FeatureInteractionEngine:
"""Production-grade feature interaction engine with validation and logging."""
def __init__(self, degree: int = 2, interaction_pairs: List[List[str]] = None):
self.degree = degree
self.interaction_pairs = interaction_pairs or []
self.poly_transformer = PolynomialFeatures(degree=degree, include_bias=False)
logger.info(f"Initialized FeatureInteractionEngine with degree={degree}")
def execute(self, data: pd.DataFrame) -> Dict[str, Any]:
"""Execute feature interaction pipeline on input data."""
try:
if data is None or data.empty:
raise ValueError("Input data cannot be None or empty")
numerical_cols = data.select_dtypes(include=[np.number]).columns.tolist()
if len(numerical_cols) < 2:
raise ValueError("Requires at least two numerical columns")
logger.info(f"Processing {len(numerical_cols)} numerical columns")
# Generate polynomial interactions
poly_features = self.poly_transformer.fit_transform(data[numerical_cols])
poly_names = self.poly_transformer.get_feature_names_out(numerical_cols)
transformed_df = pd.DataFrame(poly_features, columns=poly_names, index=data.index)
# Add custom cross-features
for pair in self.interaction_pairs:
if len(pair) == 2 and pair[0] in data.columns and pair[1] in data.columns:
col_name = f"{pair[0]}_x_{pair[1]}"
transformed_df[col_name] = data[pair[0]] * data[pair[1]]
logger.info(f"Added cross-feature: {col_name}")
result = {
'status': 'success'
'transformed_data': transformed_df
'metadata': {
'original_columns': len(data.columns)
'new_columns': len(transformed_df.columns)
'interaction_degree': self.degree
'rows_processed': len(data)
}
}
logger.info("Feature interaction completed successfully")
return result
except Exception as e:
logger.error(f"Feature interaction failed: {str(e)}")
return {'status': 'error', 'message': str(e)}
# Self-contained test
if __name__ == "__main__":
logging.basicConfig(level=logging.INFO)
test_df = pd.DataFrame({
'x1': np.random.randn(100)
'x2': np.random.randn(100)
'x3': np.random.randn(100)
})
engine = FeatureInteractionEngine(degree=2, interaction_pairs=[['x1', 'x2']])
output = engine.execute(test_df)
print(f"Status: {output['status']}, New columns: {output['metadata']['new_columns']}")
Best Practices
- ✅ Always validate your implementation on test data
- ✅ Document your assumptions and methodology
- ✅ Use version control for reproducibility
- ✅ Monitor performance metrics in production
- ✅ Periodically review and update your approach
- ✅ Test with edge cases and outliers
- ✅ Log all significant operations for debugging
Common Pitfalls
| Pitfall | Problem | Solution | |
Constraints
MUST DO
- Validate all data preprocessing steps are fit-only on training data, never on validation or test sets
- Implement reproducible pipelines with fixed random seeds and deterministic operations where possible
- Report model performance with confidence intervals via bootstrapping or cross-validation across multiple runs
- Log all experiments with parameters, metrics, and artifacts using MLflow or equivalent tracking system
MUST NOT DO
- Do not evaluate a model on the same data used for training — always hold out a proper test set
- Avoid overfitting to the validation set by limiting hyperparameter search iterations
- Never use features that can only be computed at inference time (look-ahead bias)
- Do not report single-run accuracy without statistical significance testing or error bars
Live References
Authoritative documentation links for this skill's domain. The model follows markdown links at load time to resolve external references and inline content.