Cross-Adaptation

A Python library for multi-target domain adaptation that enables robust machine learning models across different domains.

📖 Overview

Cross-adaptation performs domain adaptation across multiple training datasets {D₁, D₂, ..., Dₙ} by iteratively changing which dataset serves as the source and which as targets. This multi-target domain adaptation approach:

• Transforms datasets through iterative source-target switching
• Works with any domain adaptation method as the underlying algorithm g
• Improves generalization to new, unseen domains
• Concatenates transformed datasets D_x = {D_x1, D_x2, ..., D_xn} for final model training

The resulting model f can successfully predict on domains not present during training.

🔬 Algorithm

🚀 Quick Start

Prerequisites

• Python ≥ 3.12
• uv (recommended) or Poetry

Installation

# Using uv (recommended)
uv sync

# Or using Poetry
poetry install

Basic Usage

import pandas as pd
from adapt.instance_based import KMM
from sklearn.tree import DecisionTreeClassifier
from cross_adaptation import Adapter

# Prepare your datasets (each DataFrame should have a 'target' column)
data = {
    "dataset1": pd.DataFrame(...),  # Your first domain dataset
    "dataset2": pd.DataFrame(...)   # Your second domain dataset
}

# Choose your domain adaptation method
adapt_model = KMM(kernel="rbf", gamma=1.0)

# Select your classifier
classifier = DecisionTreeClassifier(random_state=42)

# Create and run the adapter
adapter = Adapter(
    data=data,
    adapt_model=adapt_model,
    classifier=classifier
)

# Get the adapted dataset
adapted_dataset = adapter.adapt()

🧪 Running Experiments

Configuration

Adjust experiment settings in experiments/config/:

• Modify dataset parameters
• Choose domain adaptation methods
• Set classifier options

Execution

# Using uv
uv run experiments/main.py

# Using Poetry
poetry run experiments/main.py

# Or using the provided script
./run_experiments.sh

📊 Key Features

• Multiple Domain Support: Handle any number of source/target domain combinations
• Method Agnostic: Compatible with various domain adaptation algorithms from the adapt library
• Sklearn Integration: Seamless integration with scikit-learn classifiers
• Experimental Framework: Built-in experiment management with Hydra configuration
• Visualization Tools: Analysis and visualization utilities for adaptation results

🛠️ Dependencies

Key libraries used:

• adapt: Domain adaptation algorithms
• scikit-learn: Machine learning utilities
• pandas: Data manipulation
• transformers: For advanced ML models
• wandb: Experiment tracking
• hydra-core: Configuration management

📁 Project Structure

cross-adaptation/
├── src/
│   ├── cross_adaptation/       # Main library code
│   └── visualization/          # Visualization utilities
├── experiments/               # Experimental framework
├── docs/                     # Documentation and assets
└── main.py                  # Main execution script

🤝 Contributing

Contributions are welcome! Please feel free to submit issues and enhancement requests.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🔬 Research

If you use this library in your research, please consider citing our work.

📞 Support

For questions and support, please open an issue on the GitHub repository.