fusilli

🌸 Don't be silly, use fusilli for all your multi-modal data fusion needs! 🌸

Introduction

🍝 Welcome to fusilli 🍝, the ultimate library for multi-modal data fusion in machine learning! Fusilli makes data fusion a piece of cake, providing a platform to combine different data types efficiently.

What can Fusilli do?

Multi-modal data fusion is the combination of different types of data (or data modalities) in the pursuit of some common goal. For example, using both blood test results and neuroimaging to predict whether somebody will develop a disease. There are many different ways to combine data modalities, and the aim of fusilli is to provide a platform for anybody to compare different methods against each other.

Fusilli is built using PyTorch Lightning and PyTorch Geometric, and it currently supports the following scenarios:

1. Tabular-Tabular Fusion: Combine two (or, for most methods, three) different types of tabular data.
2. Tabular-Image Fusion: Combine one type of tabular data with image data (2D or 3D).

Fusilli supports a range of prediction tasks, including regression, binary classification, and multi-class classification. Note that it does not currently support tasks such as clustering or segmentation.

Want to know more? Here is a link to Read the Docs

Installation

To savour the flavours of fusilli, you can install it using pip:

pip install fusilli

Quick Start

Here is a quick example of how to use fusilli to train a regression model and plot the real values vs. predicted values.

    from fusilli.data import prepare_fusion_data
    from fusilli.train import train_and_save_models
    from fusilli.eval import RealsVsPreds
    import matplotlib.pyplot as plt

    # Import a fusion model
    from fusilli.fusionmodels.tabularfusion.concat_data import ConcatTabularData as ExampleModel

    data_paths = {
        "tabular1": "path/to/tabular_1.csv",  
        "tabular2": "path/to/tabular_2.csv",  
        "image": "path/to/image_file.pt",  
    }

    output_paths = {
        "checkpoints": "path/to/checkpoints/dir",  
        "losses": "path/to/losses/dir",  
        "figures": "path/to/figures/dir",  
    }

    # Get the data ready
    data_module = prepare_fusion_data(prediction_task="regression",
                                      fusion_model=ExampleModel,
                                      data_paths=data_paths,
                                      output_paths=output_paths)

    # Train the model
    trained_model = train_and_save_models(data_module=data_module,
                                          fusion_model=ExampleModel)

    # Evaluate the model by plotting the real values vs. predicted values
    RealsVsPreds_figure = RealsVsPreds.from_final_val_data(trained_model)
    plt.show()

How to Cite

Florence Townend, Patrick J. Roddy, & Philipp Goebl. (2024). florencejt/fusilli: Fusilli v1.1.0 (v1.1.0). Zenodo. https://doi.org/10.5281/zenodo.10463697

Contribute!

If you've developed new fusion methods or want to enhance Fusilli, check our contribution guidelines to get started. PyTorch templates and contribution guidance our in the contributions documentation.

Authors and Acknowledgements

fusilli is authored by Florence J Townend, James Chapman, and James H Cole.

Florence J Townend was supported by a UCL UKRI Centre for Doctoral Training in AI-enabled Healthcare studentship ( EP/S021612/1) during the initial development of this software.

License

This project is licensed under AGPLv3. See the LICENSE file for details.