Amplio is an interactive research tool for data augmentation. The system visualizes the embeddings of input sentences and helps users systematically explore and fill in "empty data spaces," i.e., parts of the desired dataset distribution with few or no data points. To do this, Amplio includes a suite of three human-in-the-loop methods for augmenting unstructured text datasets: Augment with LLM, Augment by Interpolation, and Augment with Concepts.
This code accompanies the research paper:
Exploring Empty Spaces: Human-in-the-Loop Data Augmentation
Catherine Yeh, Donghao Ren, Yannick Assogba, Dominik Moritz, Fred Hohman
arXiv, 2024.
Paper: https://arxiv.org/abs/2410.01088
The system setup requires running two main components: (1) the backend server and (2) the frontend interface. The backend and frontend run on separate servers.
First, create a secrets file and install the pipenv environment.
To add your API key, duplicate secrets_example.json and rename it to secrets.json. Then copy and paste your own API keys in secrets.json.
Install pipenv:
pip install pipenv
Build requirements for project:
pipenv install
Start virtual env:
pipenv shell
Navigate into backend folder:
cd backend
Start the backend server:
python server.py
The server should now be running at 127.0.0.1:5000.
After the backend server is running, in a separate terminal window, navigate into frontend folder:
cd frontend
Install dependencies:
npm install
Start frontend development server:
npm run dev
The interface should now be live at localhost:5173.
All data needed to run the system is available in the data folder. This data was generated using Python 3.11.
Similarly, all models needed to run the system are available in the models folder.
Note: you may run into issues if your Python version != 3.11. In this case, please run the data/generate_data.ipynb notebook to regenerate the data and model files needed to run the demo. You can also use this notebook to add new datasets.
If you add a new dataset you will need to update these files:
Look for the sections marked with UPDATE HERE IF YOU ADD A NEW DATASET.
Similarly, if you want to remove a dataset from the system, you will need to edit the files above.
When making contributions, refer to the CONTRIBUTING guidelines and read the CODE OF CONDUCT.
To cite our paper, please use:
@article{yeh2024exploring,
title={{Exploring Empty Spaces: Human-in-the-Loop Data Augmentation}},
author={Yeh, Catherine and Ren, Donghao and Assogba, Yannick and Moritz, Dominik and Hohman, Fred},
journal={arXiv preprint arXiv:2410.01088},
year={2024},
doi={10.48550/arXiv.2410.01088}
}This code is released under the LICENSE terms.