HypperSteer

An Interactive Tool for Prescriptive Sequence Predictions with RNN

Table of Contents

• About the Project
  • Built With
• Getting Started
  • Prerequisites
  • Installation
• Usage
  • Quick Start
  • Demo
• License
• Contact

About The Project

Built With

• React
• deck.gl
• REST API
• PyTorch

Prerequisites

• npm

npm install npm@latest -g

• Python3

Installation

1. Clone the repo

git clone https://github.com/nauhc/hyppersteer.git

3. Install NPM packages

npm install

Usage

Quick Start

This is an example of how to set up your project locally. To get a local copy up and running follow these simple example steps, after installing all dependencies (root directory by default):

cd backend/
python3 api.py

to start the deep learning model server, and

cd (root)
yarn start

to start the web-based UI for interactions.

Demo

HypperSteer helps to explore individual data instances and their prediction results. Each dot in the 2D projection view represents an instance with its class represented by the color. Perturb any feature values at any time-steps and predict with the RNN model.

For the biLSTM model I trained, see this repo.

In the following example, we train a biLSTM model that uses electronic health records to predict patients' mortality. The following demo visualizes the health records of two patients (one dead and one alive).

Here, perturbing the patient's "joint fluid" values at the last three time-steps alters the mortality prediction result from the dead to alive!

But for a random patient, what features and what time-step to perturb for the desired result?

Our paper HypperSteer further discusses the counterfactual and partial dependence analysis for hypothetical steering.

Cite our paper if you find the source code or the paper to be helpful.

@misc{wang2020hyppersteer,
      title={HypperSteer: Hypothetical Steering and Data Perturbation in Sequence Prediction with Deep Learning},
      author={Chuan Wang and Kwan-Liu Ma},
      year={2020},
      eprint={2011.02149},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Contributing

Contributions are what make the open source community such an amazing place to be learn, inspire, and create. Any contributions you make are greatly appreciated.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

License

Distributed under the MIT License. See LICENSE for more information.

Contact

Project Link: https://github.com/nauhc/hyppersteer