Housing Price Prediction — Data Preprocessing and Custom Transformers

This Jupyter notebook contains the full walkthrough of Chapter 2 from Aurélien Géron's "Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow", focused on data preprocessing, feature engineering, and building custom transformers.

---

Contents

• Data Loading and Exploration
  Initial inspection of the housing dataset, exploration of key features and labels.

• Data Cleaning and Imputation
  Handling missing values using SimpleImputer.

• Feature Engineering
  Creating new features such as ratios (e.g., rooms_per_household, people_per_household).

• Custom Transformer: KNeighborsRegressor Wrapper
  Implemented a transformer that wraps a k-Nearest Neighbors regressor, training it on latitude and longitude to generate a new feature estimating median housing price from nearby districts.

• Custom Transformer: StandardScalerClone
  Built a custom scaler replicating StandardScaler functionality, including:

  • Fit and transform methods
  • Inverse transform support
  • Proper handling of feature names

• Pipelines and ColumnTransformers
  Combined various preprocessing steps into pipelines for numerical and categorical features. Integrated custom transformers seamlessly.

• Hyperparameter Tuning
  Discussed tuning hyperparameters of the custom KNN regressor transformer within the pipeline.

• Feature Names and Interpretability
  Added get_feature_names_out() methods to custom transformers for pipeline compatibility and clearer feature naming.

---

Key Learnings

• How to implement custom transformers by extending BaseEstimator and TransformerMixin.
• Importance of handling pandas DataFrames vs. numpy arrays in transformers.
• Usage of check_array, check_is_fitted, and managing input/output feature names.
• Constructing reusable preprocessing pipelines with Pipeline and ColumnTransformer.
• Techniques for incorporating machine learning models (like KNN) as feature generators inside transformers.
• How to expose and tune hyperparameters of nested estimators.
• Best practices for making custom transformers compatible with scikit-learn tools.

---

Usage

• Run the notebook sequentially to follow the preprocessing and feature engineering steps.
• Modify parameters in the custom transformers and pipelines to experiment with different setups.
• Use the provided code snippets to build similar custom transformers for other regressors or estimators.

---

Requirements

• Python 3.7+
• scikit-learn
• numpy
• pandas
• matplotlib (optional, for visualization)

You can install dependencies using:

pip install scikit-learn numpy pandas matplotlib