Fake news detection

How to run training

1. Install dependencies

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Set PYTHONPATH variable

export PYTHONPATH=$(pwd)

3. Download dataset from Kaggle Place them into ./dataset folder

• Fake.csv
• True.csv

4. Run training pipeline via command below

python3 script_train.py \
    --device cuda \
    --fake-path "./dataset/Fake.csv" \
    --true-path "./dataset/True.csv" \
    --cache-folder "./cache/" \
    --batch-size 32 \
    --epoches 50 \
    --last-states 1 \
    --arch deep_normalized_class_bert\
    --optim adam

Supported models

1. Classification Bert Model
2. Classification Bert Model with batch normalization
3. Classification Bert Model with batch normalization and hidden linear layer

Run script_train.py -h to get help