Deep Continual Auditing

PyTorch implementation of Continual Learning for Unsupervised Anomaly Detection in Continuous Auditing of Financial Accounting Data [link to the paper]

Code Structure

├── DeepContinualAuditing                    
    ├── BenchmarkConfigs
        ├── ...                   # benchmark config files as YAML files
    ├── Data
        ├── ...                   # Datasets as CSV files (should be copied here)
    ├── ExperimentHandler
        ├── ...                   # Implementation of different strategies (CL, Scratch, Joint)
    ├── NetworkHandler
        ├── ...                   # Implementation of the autoencoder model used in the experiments
    ├── Scripts
        ├── ...                   # Scripts for reproducibility
    ├── UtilsHandler
        ├── ...                   # Different util files for strategy and benchmark
    ├── main.py                   # Main function is implemented here.

Datasets

Datasets used in the paper can be downloaded from here:
LINK-TO-BE-ADDED
After downloading the CSV files, copy them to ./Data/ in the main directory of the repository.

Running an experiment

All scripts to reproduce results are saved under ./Scripts/, therefore to run an experiment you can simply execute the following command:

bash Scripts/FOLDER_NAME/BASH_SCRIPT_FILENAME.sh 

If datasets are stored in a different folder than ./Data, you need to change --data_dir in the script you aim to run correspondingly.