Cockpit: A Practical Debugging Tool for Training Deep Neural Networks
A Practical Debugging Tool for Training Deep Neural Networks
A better status screen for deep learning.
Installation • Docs • Experiments • License • Citation
pip install cockpit-for-pytorch
Cockpit is a visual and statistical debugger specifically designed for deep learning. Training a deep neural network is often a pain! Successfully training such a network usually requires either years of intuition or expensive parameter searches involving lots of trial and error. Traditional debuggers provide only limited help: They can find syntactical errors but not training bugs such as ill-chosen learning rates. Cockpit offers a closer, more meaningful look into the training process with multiple well-chosen instruments.
Installation
To install Cockpit simply run
pip install cockpit-for-pytorch
Conda environment
For convenience, we also provide a conda environment, which can be installed via the conda yml file. It includes all the necessary requirements to build the docs, execute the tests and run the examples.Documentation
The documentation provides a full tutorial on how to get started using Cockpit as well as a detailed documentation of its API.
Experiments
To showcase the capabilities of Cockpit we performed several experiments illustrating the usefulness of our debugging tool. The code for the experiments can be found in a separate repository. For a discussion of those experiments please refer to our paper.
License
Distributed under the MIT License. See LICENSE for more information.
Citation
If you use Cockpit, please consider citing:
@misc{schneider2021cockpit,
title={{Cockpit: A Practical Debugging Tool for Training Deep Neural Networks}},
author={Frank Schneider and Felix Dangel and Philipp Hennig},
year={2021},
eprint={2102.06604},
archivePrefix={arXiv},
primaryClass={cs.LG}
}