Skip to content

Latest commit

 

History

History
36 lines (25 loc) · 2.42 KB

problem.md

File metadata and controls

36 lines (25 loc) · 2.42 KB

MGSC 673: Final Project

Objective

Your objective in this project is to build a multi-task learning model that predicts both house prices (a regression task) and house category (a classification task). This involves understanding and implementing multi-task learning models, and using PyTorch Lightning's advanced features for managing such complex projects.

Dataset

House Prices - Advanced Regression Techniques Dataset

You will create a new variable 'House Category' based on 'House Style', 'Bldg Type', 'Year Built', and 'Year Remod/Add' for the classification task.

Steps

  • Data Exploration and Preprocessing: Understand your data. Handle missing values, encode categorical variables, and normalize numerical variables.
  • Multi-task Model Building: Use PyTorch Lightning to build a feed-forward neural network model that predicts both house prices (a regression task) and house category (a classification task). This requires a shared bottom model and task-specific top layers.
  • Activation Functions and Optimizers: Experiment with various activation functions and optimizers. Compare their effects on the performance of your model.
  • Loss Functions: Implement and use appropriate loss functions for both tasks. Combine these into a single loss function for training your multi-task model.
  • Model Evaluation: Use suitable metrics to evaluate the performance of your model on both tasks. Advanced PyTorch Lightning Features: Use PyTorch Lightning's features like logging, callback system, and Trainer API to effectively manage your project.
  • Hyperparameter Tuning: Use PyTorch Lightning's integration with Optuna for hyperparameter optimization.
  • Report: Write a comprehensive report detailing your approach, experiments, and results.

Deliverables

  1. Python scripts for data preprocessing, model building, training, and evaluation.
  2. Trained model files.
  3. A comprehensive report on your findings.

Learning Outcomes

  • Understanding and implementing multi-task learning models with PyTorch and PyTorch Lightning.
  • Applying different activation functions, optimizers, and loss functions in a multi-task learning context.
  • Using advanced features of PyTorch Lightning.
  • Performing hyperparameter tuning with PyTorch Lightning's integration with Optuna or Ray Tune.
  • Handling real-world data and predicting outcomes for multiple tasks using a shared model architecture.