UniPredict

This repo is a demo for supporting the reproducibility of the UniPredict framework. The official repository will be released after the review process finishes.

Example Usage

Python 3.11

Install Dependencies

pip install -r requirements.txt

After dependencies installed, you can choose to run either from any checkpoints we provided, or a provided pipeline. For the latter, run

./run.sh

For the former, follow the steps below:

Download & Preprocess Datasets

We provide several checkpoints for downloading and preprocessing datasets.

Start everything from scratch

python download_data.py --from-step scratch

This is not recommended as it will download > 2000 datasets to your computer. As an alternative, we provided a small subset for you to test on.

Start from metadata

python download_data.py --from-step round_1

Before running this line, you have to put a valid openai api key to .env file. You are not required to do your own metadata preprocessing because we have included the preprocessed metadata in each provded datasets.

Start from preprocessing

python download_data.py --from-step metadata

Run

python preprocess_data.py --model-type unipred
python train.py --model-type unipred

Test

python test_model.py

Plot Graphs

Use display_data.ipynb.