feature-engineering-nikel 0.1.0

A short human-friendly description of your package

Feature Engineering Suite

This is a comprehensive and flexible Python library for performing common feature engineering tasks, designed to be easily integrated into Scikit-learn pipelines.

Installation

First, ensure you have the necessary files (setup.py and the feature_engineering_suite directory) structured correctly.

Navigate to the root directory (the one containing setup.py) in your terminal and run this command to create a source distribution:

python setup.py sdist

This will create a dist directory containing a file like feature_engineering_suite-0.1.0.tar.gz. You can now install your package using pip:

pip install dist/feature_engineering_suite-0.1.0.tar.gz

How to Use

The library is designed to be intuitive and flexible. Here's a complete example.

1. Sample Data

Let's start with a sample dataset.

import pandas as pd  
import numpy as np

# Create a sample DataFrame for a classification problem  
data = {  
    'age': [25, 30, 35, 40, 45, 50, 55, 60, 65, 70],  
    'salary': [50000, 60000, 75000, 90000, 110000, 135000, 160000, 180000, 210000, 240000],  
    'years_experience': [2, 5, 8, 12, 15, 18, 22, 25, 28, 30],  
    'department': ['HR', 'IT', 'Sales', 'IT', 'Sales', 'HR', 'IT', 'Sales', 'HR', 'IT'],  
    'education': ['Bachelor', 'Master', 'Bachelor', 'PhD', 'Master', 'Bachelor', 'PhD', 'Master', 'Bachelor', 'PhD'],  
    'purchased_premium': [0, 0, 1, 0, 1, 1, 1, 0, 1, 1]  
}  
df = pd.DataFrame(data)  
df['salary_correlated'] = df['salary'] * 1.1 + np.random.normal(0, 5000, df.shape[0])

X = df.drop('purchased_premium', axis=1)  
y = df['purchased_premium']

2. Feature Selection

First, let's identify the most important and least redundant features.

from feature_engineering_suite import FeatureSelector

# Get feature importance scores  
importance = FeatureSelector.get_feature_importance(X.select_dtypes(include=np.number), y, task='classification')  
print("--- Feature Importance ---")  
print(importance)

# Find and remove highly correlated features  
corr_selector = FeatureSelector(correlation_threshold=0.9)  
corr_selector.fit(X.select_dtypes(include=np.number))  
print(f"\n--- Features to Drop (Correlation > 0.9) ---n{corr_selector.features_to_drop_}")  
X_uncorrelated = corr_selector.transform(X)  
print(f"nShape of X before dropping correlated features: {X.shape}")  
print(f"Shape of X after dropping correlated features: {X_uncorrelated.shape}")

3. Transformation and Standardization

Now, let's apply transformations to the numerical features.

from feature_engineering_suite import Standardizer, LogTransformer

# Apply standard scaling to 'age' and 'years_experience'  
standardizer = Standardizer(columns=['age', 'years_experience'])  
X_scaled = standardizer.fit_transform(X_uncorrelated)

# Apply log transformation to the 'salary' column  
log_transformer = LogTransformer(columns=['salary'])  
X_final_numeric = log_transformer.fit_transform(X_scaled)

print("\n--- Data After Transformations ---")  
print(X_final_numeric.head())

4. Categorical Encoding

Finally, let's encode the categorical features.

from feature_engineering_suite import Encoder

# Define an ordinal mapping for the 'education' column  
education_map = {'Bachelor': 1, 'Master': 2, 'PhD': 3}

# Use the Encoder for both one-hot and ordinal encoding  
# We will one-hot encode 'department' and ordinally encode 'education'

# One-hot encode department  
onehot_encoder = Encoder(method='onehot', columns=['department'])  
X_encoded = onehot_encoder.fit_transform(X_final_numeric)

# Ordinal encode education  
ordinal_encoder = Encoder(method='ordinal', columns=['education'], mapping={'education': education_map})  
X_fully_processed = ordinal_encoder.fit_transform(X_encoded)

print("\n--- Fully Processed DataFrame ---")  
print(X_fully_processed.head())  
print(f"\nFinal shape of processed data: {X_fully_processed.shape}")

This library provides the building blocks you need to create powerful and reproducible feature engineering pipelines for any dataset.