BoostCD

1. Setup

Start by cloning the repository:

git clone https://github.com/epfl-dlab/BoostCD.git

We recommend creating a new conda virtual environment as follows:

conda env create -f environment.yml

This command also installs all the necessary packages.

2. Downloading data and models

The data is available on huggingface and can be loaded with

from datasets import load_dataset
dataset = load_dataset("msakota/boostie")

3. Usage

Training

To train a model from scratch on the desired data, run:

# specify a directory where training data is located
RUN_NAME="train_boostie_base_fe"
python src/genie/run_train.py run_name=$RUN_NAME +experiment/train=boostie_base_fe

Inference

To run inference on a trained model:

CHECKPOINT_PATH="./models/boostie_base_fe.ckpt" # specify path to the trained model
RUN_NAME="inference_boostie_base_fe"
python src/genie/run_inference.py run_name=$RUN_NAME checkpoint_path=$CHECKPOINT_PATH +experiment/inference=boostie_base_fe

Evaluation

To compute the micro and macro performance, as well as the performance bucketed by relation frequency and number of target triplets, you only need the run's WandB path and to execute:

python src/genie/run_process_predictions.py +experiment/process_predictions=complete_boostie wandb_run_path=$WANDB_PATH

Citation

If you found our resources useful, please consider citing our work.

@misc{šakota2025combiningconstrainedunconstraineddecoding,
      title={Combining Constrained and Unconstrained Decoding via Boosting: BoostCD and Its Application to Information Extraction}, 
      author={Marija Šakota and Robert West},
      year={2025},
      eprint={2506.14901},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.14901}, 
}