Skip to content

Welcome to Framework3

A Modular and Extensible Framework for ML Experimentation

Framework3 is a lightweight, flexible foundation for building machine learning experimentation systems that are structured, extensible, and reusable. Unlike rigid monolithic platforms, Framework3 lets you craft your own workflows by composing modular components or reusing existing ones.

Its architecture is grounded in principles of modularity, composability, and transparency, enabling you to tailor each part of your workflow to fit your project's needs.

Warning

Framework3 is under active development. While we strive for stability, caution is advised when using it in production environments.

What is Framework3?

Framework3 is not a closed pipeline, nor a library of pre-trained models. It’s a collection of base interfaces and pluggable components that empower you to:

  • Design and combine filters, metrics, optimizers, and storers.
  • Build reusable pipelines — sequential or parallel.
  • Execute optimization or evaluation processes, even in distributed settings.
  • Structure your experiments with clarity and complete control.

Think of Framework3 as a toolbox of interoperable building blocks — not a black-box solution.

Key Features

  • 🧩 Modular architecture: easily combine and reuse components.
  • 🔌 Extensible plugin system: add new filters, metrics, or storers without modifying the core.
  • 🧪 Composable pipelines: design sequential, parallel, or hybrid flows.
  • 🧠 Built-in optimization and evaluation: test different configurations and track performance.
  • 🧵 Distributed processing: supports parallel execution and MapReduce-style pipelines.

Design Philosophy

  1. Composition over configuration: each piece is designed to work seamlessly with others.
  2. Frictionless extensibility: implement your own components by subclassing base interfaces.
  3. Full transparency: inspect every step, metric, and decision in the workflow.

Conceptual Diagram

graph TD
    subgraph Core
        A[BaseFilter]
        B[BasePipeline]
        C[BaseOptimizer]
        D[BaseSplitter]
        E[BaseMetric]
        F[BaseStorer]
        A --> B
        B --> C
        B --> D
        B --> E
        B --> F
    end

    subgraph Plugins
        G[F3Pipeline] --> B
        H[KFoldSplitter] --> D
        I[OptunaOptimizer] --> C
        J[LocalStorer] --> F
        K[KnnFilter] --> A
        L[F1Metric] --> E
        M[ParallelPipeline] --> B
        N[RemoteStorer] --> F
    end

Repository Structure

Framework3 is structured to be easy to navigate, extend, and maintain:

  • base/: Fundamental interfaces and abstractions.
  • plugins/: Filters, pipelines, metrics, optimizers, storers, and more.
  • container/: Dependency injection system.
  • examples/: Real-world use cases and tutorials.
  • tests/: Unit and integration tests.
  • docs/: Full documentation, including this page.

🚀 Ready to Get Started?

🤝 Contribute

Interested in contributing? Check out the Contribution Guidelines.

📄 License

Framework3 is licensed under AGPL-3.0 — View License