rstudio/keras3

R Interface to Keras

What it solves

It provides an R interface to Keras, allowing R users to build and train deep learning models using a high-level, user-friendly API designed for fast experimentation.

How it works

The package acts as an interface to the Keras API, enabling the creation of complex neural network architectures. It supports seamless execution on both CPUs and GPUs, and provides built-in support for various network types, including convolutional networks for computer vision and recurrent networks for sequence processing.

Who it’s for

Data scientists and researchers using R who want to prototype and deploy deep learning models quickly without needing to switch to Python.

Highlights

  • Seamless CPU and GPU execution
  • Support for convolutional and recurrent networks
  • Ability to build arbitrary network architectures, including multi-input and multi-output models
  • Focus on reducing the delay between idea and result

Related

  • Project
    keras-team/keras

    A multi-backend deep learning framework that allows users to build and train models across JAX, TensorFlow, and PyTorch to avoid framework lock-in and optimize performance.

  • Project
    topepo/caret

    A comprehensive R package for training and plotting classification and regression models, providing a unified interface for machine learning tasks.

  • Project
    keras-team/keras-hub

    KerasHub is a pretrained modeling library that provides multi-framework implementations of popular AI models for text, image, and audio, supporting JAX, TensorFlow, and PyTorch.

  • Dispatch
    Llama 3.2 in Keras

    Llama 3.2 is fully supported in Keras via keras-hub, allowing users to load Hugging Face checkpoints and run models across JAX, PyTorch, or TensorFlow backends.

  • Project
    tidymodels/parsnip

    parsnip is an R package that provides a tidy, unified interface for specifying and fitting statistical and machine‑learning models. It lets you describe a model (type, mode, hyper‑parameters) once and then choose any supported engine (e.g., ranger, spark) to train it, harmonising argument names and simplifying code across different packages.