/aac_anomaly

Streamlit dashboard for time series anomaly detection

Primary LanguageJupyter Notebook

Dashboard for anomaly detection in seasonal time series

An ensemble of different time series anomaly detection methods is being constructed to estimate the probability of an anomaly at time $t$. Preprocessing is done via (hierarchically) aggregating the data to univariate conditional time series. If certain criteria are not met (e.g. minimum sample size), then a particular time series is further aggregated (i.e. aggregation follows a binary-tree).

Package structure

├── app.py
├── docker-compose.yaml
├── Dockerfile
├── environment.yml
├── README.md
├── requirements.txt
├── run_app_locally.sh
└── src
    ├── aac_ts_anomaly
    │   ├── config
    │   │   ├── aws_config.py
    │   │   ├── global_config.py
    │   │   ├── __init__.py
    │   ├── data
    │   ├── resources
    │   ├── services
    │   │   ├── file_aws.py
    │   │   ├── file.py
    │   │   ├── __init__.py
    │   └── utils
    ├── __init__.py
    ├── notebooks
    ├── setup.py
    └── templates

Create conda environment with require packages installed

#conda env create -f environment.yml   # optionally
conda create -n env_tsanomaly
conda activate env_tsanomaly

To install the package locally, execute the following steps:

cd aac_ts_anomaly
pip install -r requirements.txt         # in case no environment.yml was used
pip install -e src

Start streamlit application by running:

bash run_app_locally.sh

Build image and start container:

docker-compose up -d