# An Unsupervised Baseline For Dialogue Breakdown Detection\\ Using Ouf-of-distribution Detection


## Instructions

Python 3.8 is used to run the scripts. We recommend using a virtual environment like (Ana|Mini)conda. Steps:

1. Setup virtual environment & install dependencies
    - `conda env create -f environment.yml`
2. [OPTIONAL] Download training data & train all models
    - `bash run_training_all.sh`
    - WARNING: executing this script might take several days of time.
        - It first downloads and prepares the DSTC8 training. Depending on the internet connection and amount CPU & RAM, this might take up to a 2-3 days.
        - Trains all 3 OoD models.
    - The step is optional since model checkpoints are already included.
3. Run evaluation on DBDC4 English data & calculate metrics:
    - `bash run_eval_all.sh`