Dockerized version of the winning submission to ValueEval'23 by Schroter et al. [original code]
Needs a Docker installation.
The classifier requires two mounted folders, input
(containing an arguments.tsv
from the dataset) and output
.
mkdir input output
curl https://zenodo.org/record/7550385/files/arguments-test.tsv > input/arguments.tsv
Create a submission file in output
:
docker run --rm -it \
--volume "$PWD/input:/input" \
--volume "$PWD/output:/output" \
ghcr.io/webis-de/valueeval23-adam-smith-12:1.0.0-cpu \
python3 /app/predict.py --inputDataset /input --outputDir /output
This repository is based on the predict.ipynb
as well as the data_modules
, models
, and toolbox
directories of https://github.com/danielschroter/human_value_detector. The parts of the code copied from the predict.ipynb
are marked with START
and END
. Necessary changes are marked in the code.
Unzip the model files into checkpoints (creating checkpoints/human_value_trained_models
with 24 files inside).
docker build -t ghcr.io/webis-de/valueeval23-adam-smith-12:1.0.0-cpu -f Dockerfile .