UCCA 4 BPM
Minimal Installation for Reproducing Our Results
install Miniconda: https://docs.conda.io/en/latest/miniconda.html
create new venv with conda create --name ucca4bpm and activate it conda activate ucca4bpm
run conda config --add channels anaconda
run conda config --add channels conda-forge
run conda install python=3.6.11 gensim=3.8.3 matplotlib=3.3.3 scikit-learn=0.20.1
run pip install dgl==0.5.2 tensorflow==2.3.1
Make sure to edit the dgl config file to use tensorflow instead of pytorch, see https://docs.dgl.ai/en/0.4.x/install/backend.html
Reproducing Our Results
Download the prepared data and copy them into ucca4bpm/data/transformed, so that e.g. file ours_qian_srl_google300.pickle is in ucca4bpm/data/transformed. https://drive.google.com/drive/folders/1jumtVxkOAswTOktmTxQ1ode33kf3MuTV?usp=sharing
We have prepared several scripts for generating our runs and plots.
ucca4bpm/experiments_ours.py will run the analysis of hyper parameters and features.
ucca4bpm/experiments_qian.py will run our model on the data by Qian et al and on our dataset annotated by MGTC schema.
ucca4bpm/experiments_quishpi.py will run our model on the data by Qian et al and on our dataset annotated by ATDP schema.
ucca4bpm/visualize_experiments.py will run the visualization of our analysis runs.
ucca4bpm/report.py will print the reported runs to console.
You can run all of those by python -m ucca4bpm.<your_script> when you are in the top level directory.
If you want to run the data processing yourself (you don't have to for reproducing the results), you can run the data/transform script. For detailed installation instructions you can sed a mail to julian.neuberger(at)uni-bayreuth.de or lars.ackermann(at)uni-bayreuth.de.