[논문 리뷰] giotto-tda: A Topological Data Analysis Toolkit for Machine Learning and Data Exploration

Guillaume Tauzin, Umberto Lupo외 7명|arXiv (Cornell University)|2020. 04. 06.

Topological and Geometric Data Analysis인용 수 99

한 줄 요약

giotto-tda는 scikit-learn–호환 API를 갖춘 머신 러닝에 토폴로지 데이터 분석을 도입하는 파이썬 라이브러리로, 지속형 동형성(persistent homology), Mapper, 그리고 다양한 데이터 유형에 걸친 대화형 시각화를 가능하게 합니다.

ABSTRACT

We introduce giotto-tda, a Python library that integrates high-performance topological data analysis with machine learning via a scikit-learn-compatible API and state-of-the-art C++ implementations. The library's ability to handle various types of data is rooted in a wide range of preprocessing techniques, and its strong focus on data exploration and interpretability is aided by an intuitive plotting API. Source code, binaries, examples, and documentation can be found at https://github.com/giotto-ai/giotto-tda.

연구 동기 및 목표

• Make TDA accessible to the Python data science community and researchers.
• Enable end-to-end ML pipelines that use topological features from diverse data types.
• Provide a scikit-learn–compatible interface for hyperparameter tuning, cross-validation, and feature selection in TDA.
• Offer interactive visualization and interpretation of topological signatures to aid data exploration.

제안 방법

• Provide scikit-learn–compatible estimators and transformers for transforming data into persistent diagrams and related features.
• Offer a TransformerResamplerMixin and an extended Pipeline to support time-series embedding and flexible preprocessing.
• Integrate high-performance C++ libraries via pybind11 bindings for scalable computation of persistent homology and related representations.
• Include a broad suite of persistent homology and Mapper algorithms, with support for multiple input data types (time series, images, graphs, point clouds, etc.).
• Expose plotting APIs (Plotly-based) for interactive exploration of topological features and intermediate results.
• Support deep learning estimators in pipelines through wrappers to PyTorch/TensorFlow ecosystems (e.g., skorch).
• Utilize parallelization (joblib) and memory caching to accelerate hyperparameter tuning and interactive visualization.

실험 결과

연구 질문

• RQ1 How can topological data analysis be integrated into scalable ML pipelines using a Python API compatible with scikit-learn?
• RQ2 What range of data types can be transformed into topological features via TDA preprocessing in a unified library?
• RQ3 How can persistent diagrams and their representations be used as features in end-to-end ML workflows?
• RQ4 What tooling is needed to visualize and interpret topological signatures interactively during model development?
• RQ5 How do modern TDA libraries compare in functionality and performance, and where does giotto-tda fit?

주요 결과

• giotto-tda provides a broad, scikit-learn–compatible interface for persistent homology and Mapper, enabling end-to-end pipelines.
• The library targets multiple data modalities (time series, images, graphs, point clouds) through extensive preprocessing transformers.
• It integrates state-of-the-art C++ implementations via pybind11 for performance and offers parallelized computation and memory caching.
• An interactive Plotly-based plotting API facilitates visualization and exploration of intermediate TDA results within pipelines.
• The project emphasizes documentation, tutorials, and community-driven development with CI/CD and high test coverage (98%).
• giotto-tda positions itself as a comprehensive Python library for topological machine learning and data exploration, augmenting traditional ML workflows.