🆚 argmining24-deepct-lexical-argument-retrieval

Code and data for the paper "DeepCT-enhanced Lexical Argument Retrieval".

Installation

Install Python 3.9 or higher.

Create and activate the virtual environment:

python3.9 -m venv venv/
source venv/bin/activate

Install dependencies:
```
pip install -e .
```

Usage

Follow these steps to prepare DeepCT and train it on a Slurm cluster.

Prepare DeepCT Image in Slurm

First, create an enroot image:

srun -c 4  --mem=100G --container-image=nvcr.io#nvidia/tensorflow:20.06-tf1-py3 --container-name=DeepCT --pty echo "Image created sucessfully"

In this image and using Slurm, clone the DeepCT repository:

srun -c 4  --mem=100G --container-name=DeepCT --container-writable --pty git clone https://github.com/AdeDZY/DeepCT.git

Download BERT:

srun -c 4  --mem=100G --container-name=DeepCT --container-writable --pty bash -c 'wget https://storage.googleapis.com/bert_models/2020_02_20/uncased_L-12_H-768_A-12.zip && mkdir bert-uncased_L-12_H-768_A-12 && unzip uncased_L-12_H-768_A-12.zip -d bert-uncased_L-12_H-768_A-12 && rm -f uncased_L-12_H-768_A-12.zip'

Train DeepCT with Slurm

After executing the steps to prepare the DeepCT image in Slurm, train the model with the following commands.

This guide assumes that you have the following files/directories:

<PATH-TO-TRAIN-DOCTERM-RECALL>: the training data file.
<OUTPUT-DIRECTORY>: directory where the trained model will be stored.

Train the model with the following command:

srun \
    --gres gpu:ampere:1 -c 4  --mem=100G \
    --container-mounts=<PATH-TO-TRAIN-DOCTERM-RECALL>,<OUTPUT-DIRECTORY> \
    --chdir /workspace/DeepCT \
    --container-name=DeepCT --pty \
    python3 run_deepct.py \
        --task_name=marcodoc \
        --do_train=true \
        --do_eval=false \
        --do_predict=false \
        --data_dir=<PATH-TO-TRAIN-DOCTERM-RECALL> \
        --vocab_file=/workspace/bert-uncased_L-12_H-768_A-12/vocab.txt \
        --bert_config_file=/workspace/bert-uncased_L-12_H-768_A-12/bert_config.json \
        --init_checkpoint=/workspace/bert-uncased_L-12_H-768_A-12/bert_model.ckpt \
        --max_seq_length=128 \
        --train_batch_size=16 \
        --learning_rate=2e-5 \
        --num_train_epochs=3.0 \
        --recall_field=title \
        --output_dir=<OUTPUT-DIRECTORY>

Train DeepCT with Slurm (in case `--chdir` does not work)

First, run the following command in order to start an interactive shell in the container:

srun \
    --gres gpu:ampere:1 -c 4  --mem=100G \
    --container-mounts=<PATH-TO-TRAIN-DOCTERM-RECALL>,<OUTPUT-DIRECTORY> \
    --container-name=DeepCT --pty \
    bash

Then cd to the downloaded Git repository and run the rest of the original script:

python3 run_deepct.py \
    --task_name=marcodoc \
    --do_train=true \
    --do_eval=false \
    --do_predict=false \
    --data_dir=<PATH-TO-TRAIN-DOCTERM-RECALL> \
    --vocab_file=/workspace/bert-uncased_L-12_H-768_A-12/vocab.txt \
    --bert_config_file=/workspace/bert-uncased_L-12_H-768_A-12/bert_config.json \
    --init_checkpoint=/workspace/bert-uncased_L-12_H-768_A-12/bert_model.ckpt \
    --max_seq_length=128 \
    --train_batch_size=16 \
    --learning_rate=2e-5 \
    --num_train_epochs=3.0 \
    --recall_field=title \
    --output_dir=<OUTPUT-DIRECTORY>

Download Touché runs

These commands will download the runs from the Touché 2020 task 1.

wget -c https://zenodo.org/records/6873564/files/touche2020-task1-runs-args-me-corpus-version-2020-04-01.zip
unzip touche2020-task1-runs-args-me-corpus-version-2020-04-01.zip
rm touche2020-task1-runs-args-me-corpus-version-2020-04-01.zip
mkdir -p data/runs_touche/
mv touche2020-task1-runs-args-me-corpus-version-2020-04-01/ data/runs_touche/2020/

These commands will download the runs from the Touché 2021 task 1.

wget -c https://zenodo.org/records/6873566/files/touche2021-task1-runs.zip
unzip touche2021-task1-runs.zip
rm touche2021-task1-runs.zip
mkdir -p data/runs_touche/
mv touche2021-task1-runs/ data/runs_touche/2021/

License

The source code in this repository is licensed under the MIT License.

webis-de/argmining24-deepct-lexical-argument-retrieval