Source code of our visual analytics system for the interpretation of hidden states in recurrent neural networks. This project contains source code for preprocessing data and the visual analytics system. Additionally, we added precomputed data for immediate use in the visual analysis system.

The sub directories contain the following:

dataPreparation: Python scripts to prepare data for analysis. In these scripts, Long Short-Term Memory (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.

We provide two scripts to generate data for analysis in our visual analytics system: for the IMDB and Reuters dataset as available in Keras. The output files can then be loaded into our visual analytics system; their locations have to be specified in userData.toml of the visual analytics system.

The output file of our data preparation scripts or the ones provided for demonstration can be loaded in our visual analytics system for visualization and analysis. Since we provide input files, you do not have to run the preprocessing steps and can use our visual analytics system immediately.

Please have a look at the respective README-files for more details.

IMDB and Reuters datasets: F. Chollet. Keras. GitHub. https://github.com/fchollet/keras. 2015.

You may find the most recent version of the source code on GitHub: https://github.com/MunzT/hiddenStatesVis