team-adam-smith23

Dockerized version of the winning submission to ValueEval'23 by Schroter et al. [original code]

Needs a Docker installation.

Usage

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

Provenance

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.

Build Docker Image

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 .