Visual Analytics System for Hidden States in Recurrent Neural Networks
Visual analytics system for the analysis of hidden states in recurrent neural networks.
You can find more information about our approach in the publication A Visual Analytics Tool for the Interpretation of Hidden States in Recurrent Neural Networks (see below).
This project contains source code for pre-processing IMDB and Reuters data available in Keras [1] and the visual analytics system itself. Additionally, we added precomputed data for immediate use in the visual analytics system.
The sub directories contain the following:
-
dataPreparation: Python scripts to prepare data for analysis. In these scripts, LSTM models are trained and data for our visual analytics system is exported. -
visualAnalytics: The source code of our visual analytics system to explore hidden states. -
demonstrationData: Data files for the use with our visual analytics system. The same data can also be generated with the data preparation scripts.
Our system uses the packages listed in the requirements files in each subdirectory. For the visual analytics system D3.js is used for the visualizations.
Our project is licensed under the MIT License.
When referencing our work, please cite our paper A Visual Analytics Tool for the Interpretation of Hidden States in Recurrent Neural Networks.
R. Garcia, T. Munz, and D. Weiskopf. Visual Analytics Tool for the Interpretation of Hidden States in Recurrent Neural Networks. Visual Computing for Industry, Biomedicine, and Art (VCIBA). 2021. DOI: 10.1186/s42492-021-00090-0.
@article{vciba2021hiddenStates,
author = {Garcia, Rafael and Munz, Tanja and and Weiskopf, Daniel},
title = {Visual Analytics Tool for the Interpretation of Hidden States in Recurrent Neural Networks},
journal = {Visual Computing for Industry, Biomedicine, and Art (VCIBA)},
year = {2021},
doi = {10.1186/s42492-021-00090-0},
}
[1] Chollet, F.: Keras. GitHub.https://github.com/fchollet/keras (2015)