Detecting Anomalies with Autoencoders on Data Streams

This repository contains the results for our ECML 2022 submission "Detecting Anomalies with Autoencoders on Data Streams"

Installation

git clone https://github.com/lucasczz/DAADS.git

python3 -m venv daads_env

source daads_env/bin/activate

cd DAADS

pip install -r requirements.txt

Reproducing the results

To run all experiments at once, run the run_exps.sh script located in ./scripts by

./scripts/run_exps.sh

The experiment results are stored in ./results.

Reproducing the results step by step

All experiment scripts are located in ./tools.

Evaluate all models

python ./tools/benchmark_exp.py

Run contamination experiment

python ./tools/contamination_exp.py

Run capacity experiment

python ./tools/capacity_exp.py

Run learning rate experiment

python ./tools/lr_exp.py

Obtain anomaly scores

python ./tools/scores_exp.py

Access datasets

from IncrementalTorch.datasets import Covertype, Shuttle
from river.datasets import CreditCard