Welcome to BOAT-Torch Documentation

BOAT is a compositional, gradient-based Bi-Level Optimization (BLO) Python library that focuses on abstracting the key BLO process into modular, flexible components. It enables researchers and developers to tackle learning tasks with hierarchical nested nature by providing customizable and diverse operator decomposition, encapsulation, and combination. BOAT supports specialized optimization strategies, including second-order or first-order, nested or non-nested, and with or without theoretical guarantees, catering to various levels of complexity.

In this section, we explain the core components of BOAT, how to install the PyTorch version, and how to use it for your optimization tasks. The main contents are organized as follows.

Installation Guide:

• BOAT Introduction
  • 🔑 Key Features
  • 🚀 Why BOAT?
  • 🌍 Applications
  • 🚩 Related Operations
  • 📜 License
• Installation and Usage Guide
  • 🔨 Installation
  • ⚡ How to Use BOAT
• BOAT-Torch Structure
  • Core Problem Class
  • Main Subpackages
  • Extension with Operation_Registry

Running Example

The running examples of data hypercleaning, meta learning and l2 regularization are organized as follows.

Example:

• Data HyperCleaning
  • Step 1: Data Preparation
  • Step 2: Model Definition
  • Step 3: Optimizer and Initialization
  • Step 4: Configuration Loading
  • Step 5: Main Function
• L2 Regularization
  • Step 1: Imports and Environment Setup
  • Step 2: Configuration Loading
  • Step 3: Argument Parsing and Reproducibility
  • Step 4: Data Preparation
  • Step 5: Model Definition
  • Step 6: Optimizer Setup
  • Step 7: BOAT Problem Construction
  • Step 8: Optimization Loop
• Meta-Learning
  • Step 1: Importing Libraries and Dependencies
  • Step 2: Argument and Dataset Setup
  • Step 3: Model and Optimizer Setup
  • Step 4: DataLoader Construction
  • Step 5: Bi-Level Optimization Configuration
  • Step 6: Meta-Training Loop

Indices and tables

• Index

• Module Index

• Search Page