Skip to content

Latest commit

 

History

History
108 lines (88 loc) · 4.1 KB

README.md

File metadata and controls

108 lines (88 loc) · 4.1 KB

DeepDebugger: Touch what you image

Official source code for ESEC/FSE 2023 Paper: DeepDebugger: An Interactive Time-Travelling Debugging Approach for Deep Classifiers

Paper PDF

Table of Contents

Installation

Dependencies

Please run the following commands in the command line:

$ conda create -n dd python=3.8
$ git clone https://github.com/xianglinyang/DeepDebugger.git
$ pip -r install requirements.txt

To install torch, please check link.

Usage

Quick Start

A tutoial in jupyter notebook is here.

Generally, we follow the following process to create meaningful visualization:

  1. Prepare data according to our format.

    Put all data under a folder content_path = /path/to/data

  2. Visualize the embedding Choose a visualization strategy (e.g., DVI, TimeVis or even your own visualization method!)

    • Use wrapped func from Strategy class 
      #--------DVI--------
VIS_METHOD = "DVI"
dvi_config = config[VIS_METHOD]
dvi = DeepVisualInsight(CONTENT_PATH, dvi_config)
dvi.visualize_embedding()

#--------TimeVis--------
VIS_METHOD = "TimeVis"
timevis_config = config[VIS_METHOD]
timevis = TimeVis(CONTENT_PATH, timevis_config)
timevis.visualize_embedding()
    • Directly call the module. 
      $ python dvi_main.py --content_path path/to/data
$ python timevis_main.py --content_path path/to/data

  3. Play with embedding visualization with our frontend or backend visualizer. [repo][tutorial]

  4. (optional) Design your own visualization method. [tutorial]

Full instructions

Please see our wiki for more details.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contact

If you have any questions, please feel free to reach out to me at [email protected].

Reproducibility

Follow batch_run.py run.sh to run the codes and reproduce the published results.

python batch_run.py

Citation

If you find our tool helpful, please cite the following paper:

@inproceedings{yang2023deepdebugger,
  title={DeepDebugger: An Interactive Time-Travelling Debugging Approach for Deep Classifiers},
  author={Yang, Xianglin and Lin, Yun and Zhang, Yifan and Huang, Linpeng and Dong, Jin Song and Mei, Hong},
  booktitle={Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering},
  pages={973--985},
  year={2023}
},
@inproceedings{yang2022temporality,
  title={Temporality Spatialization: A Scalable and Faithful Time-Travelling Visualization for Deep Classifier Training},
  author={Yang, Xianglin and Lin, Yun and Liu, Ruofan and Dong, Jin Song},
  booktitle = {Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, {IJCAI-22}},
  year={2022}
},
@inproceedings{yang2022deepvisualinsight,
  title={DeepVisualInsight: Time-Travelling Visualization for Spatio-Temporal Causality of Deep Classification Training},
  author={Yang, Xianglin and Lin, Yun and Liu, Ruofan and He, Zhenfeng and Wang, Chao and Dong, Jin Song and Mei, Hong},
  booktitle = {The Thirty-Sixth AAAI Conference on Artificial Intelligence (AAAI)},
  year={2022}
}