.. image:: https://img.shields.io/pypi/status/Renate :target: # :alt: PyPI - Status .. image:: https://img.shields.io/github/v/release/awslabs/Renate :target: https://github.com/awslabs/Renate/releases/tag/v0.5.1 :alt: Latest Release .. image:: https://img.shields.io/pypi/dm/Renate :target: https://pypistats.org/packages/renate :alt: PyPI - Downloads .. image:: https://img.shields.io/github/license/awslabs/Renate :target: https://github.com/awslabs/Renate/blob/main/LICENSE :alt: License .. image:: https://readthedocs.org/projects/renate/badge/?version=latest :target: https://renate.readthedocs.io :alt: Documentation Status .. image:: https://raw.githubusercontent.com/awslabs/Renate/python-coverage-comment-action-data/badge.svg :target: https://htmlpreview.github.io/?https://github.com/awslabs/Renate/blob/python-coverage-comment-action-data/htmlcov/index.html :alt: Coverage Badge

Renate: Automatic Neural Networks Retraining and Continual Learning in Python

Renate is a Python package for automatic retraining of neural networks models. It uses advanced Continual Learning and Lifelong Learning algorithms to achieve this purpose. The implementation is based on PyTorch <https://pytorch.org> and Lightning <https://www.pytorchlightning.ai> for deep learning, and Syne Tune <https://github.com/awslabs/syne-tune>_ for hyperparameter optimization.

Quick links

Install renate with pip install renate or look at these instructions <https://renate.readthedocs.io/en/latest/getting_started/install.html>_
Examples for local training <https://renate.readthedocs.io/en/latest/examples/train_mlp_locally.html>_ and training on Amazon SageMaker <https://renate.readthedocs.io/en/latest/examples/train_classifier_sagemaker.html>_.
Documentation <https://renate.readthedocs.io>_
Supported Algorithms <https://renate.readthedocs.io/en/latest/getting_started/supported_algorithms.html>_

Who needs Renate?

In many applications data is made available over time and retraining from scratch for every new batch of data is prohibitively expensive. In these cases, we would like to use the new batch of data provided to update our previous model with limited costs. Unfortunately, since data in different chunks is not sampled according to the same distribution, just fine-tuning the old model creates problems like catastrophic forgetting. The algorithms in Renate help mitigating the negative impact of forgetting and increase the model performance overall.

.. figure:: https://raw.githubusercontent.com/awslabs/Renate/main/doc/_images/improvement_renate.svg :align: center :alt: Renate vs Model Fine-Tuning.

Renate's update mechanisms improve over naive fine-tuning approaches. [#]_

Renate also offers hyperparameter optimization (HPO), a functionality that can heavily impact the performance of the model when continuously updated. To do so, Renate employs Syne Tune <https://github.com/awslabs/syne-tune>_ under the hood, and can offer advanced HPO methods such multi-fidelity algorithms (ASHA) and transfer learning algorithms (useful for speeding up the retuning).

.. figure:: https://raw.githubusercontent.com/awslabs/Renate/main/doc/_images/improvement_tuning.svg :align: center :alt: Impact of HPO on Renate's Updating Algorithms.

Renate will benefit from hyperparameter tuning compared to Renate with default settings. [#]_

Key features

Easy to scale and run in the cloud
Designed for real-world retraining pipelines
Advanced HPO functionalities available out-of-the-box
Open for experimentation

Resources

(blog) Automatically retrain neural networks with Renate <https://aws.amazon.com/blogs/machine-learning/automatically-retrain-neural-networks-with-renate/>_
(paper) Renate: A Library for Real-World Continual Learning <https://arxiv.org/abs/2304.12067>_

Cite Renate

.. code-block:: bibtex

  @misc{renate2023,
    title           = {Renate: A Library for Real-World Continual Learning}, 
    author          = {Martin Wistuba and
                       Martin Ferianc and
                       Lukas Balles and
                       Cedric Archambeau and
                       Giovanni Zappella},
    year            = {2023},
    eprint          = {2304.12067},
    archivePrefix   = {arXiv},
    primaryClass    = {cs.LG}
  }

What are you looking for?

Installation Instructions <https://renate.readthedocs.io/en/latest/getting_started/install.html>_ .. code-block:: bash

pip install renate
Examples:
- Train an MLP locally on MNIST <https://renate.readthedocs.io/en/latest/examples/train_mlp_locally.html>_
- Train a ResNet on SageMaker <https://renate.readthedocs.io/en/latest/examples/train_classifier_sagemaker.html>_
Documentation website with API doc and examples <https://renate.readthedocs.io>_
List of the supported algorithms <https://renate.readthedocs.io/en/latest/getting_started/supported_algorithms.html>_
How to run continual learning experiments using Renate <https://renate.readthedocs.io/en/latest/benchmarking/index.html>_
Guidelines for Contributors <https://github.com/awslabs/renate/tree/master/CONTRIBUTING.md>_

If you did not find what you were looking for, open an issue <https://github.com/awslabs/Renate/issues/new>_ and we will do our best to improve the documentation.

.. [#] To create this plot, we simulated domain-incremental learning with CLEAR-100 <https://clear-benchmark.github.io/>_. The training data was divided by year, and we trained sequentially on them. Fine-tuning refers to the strategy to learn on the first partition from scratch, and train on each of the subsequent partitions for few epochs only. We compare to Experience Replay with an infinite memory size. For both methods we use the same amount of training time and choose the best checkpoint using a validation set. Results reported are on the test set.

.. [#] In this experiment, we consider class-incremental learning on CIFAR-10. We compare Experience Replay against a version in which its hyperparameters were tuned.

awslabs / Renate