Real-time anomaly detection with superexperts

The repository reproduces the results of the paper Real-time anomaly detection with superexperts. If you want to see the results of the paper there is no need to run anything: the visualisation of results is available here and the analysis of losses and classification metrics is available here.

If you want to run the project the easiest way is to use Binder. Click on .

On Binder you need to first run main.ipynb to calculate the predictions of Fixed-share and Variable-share and output the results (this will take around 35 minutes on Binder or around 15 minutes to run locally). After that you can run results_analysis.ipynb and results_plots.ipynb.

If you want to run the project locally follow the installation instructions below.

Installation

Install anaconda or miniconda https://docs.conda.io/en/latest/miniconda.html.

Clone the repository (note that the flag --recursive is important as the repository contains the submodule NAB):

git clone --recursive https://github.com/alan-turing-institute/anomaly_with_experts.git anomaly_with_experts
cd anomaly_with_experts

Create a conda environment:

conda create -n anomaly_with_experts python=3.7

Activate the environment:

conda activate anomaly_with_experts

Use the package manager pip to install the requirements:

pip install -r requirements.txt --use-feature=2020-resolver

If you do not have Jupyter Notebook installed:

pip install notebook

To launch it:

jupyter notebook

After that you should be able to run main.ipynb which calculates the predictions of Fixed-share and Variable-share on NAB and outputs the results. Then you can run results_analysis.ipynb to analyse the losses and classification metrics and results_plots.ipynb to visualise the plots from the paper.