Maat

Here is the repo for our paper ``Maat: Performance Metric Anomaly Anticipation for Cloud Services with Conditional Diffusion'', accepted by ASE 2023.

Architecture

Overview

Diffusion Model

Model M

Data

The data should be stored in csv files with the first column being timestamp'' and the last column being label''. If labels are not avaliable, it can be all zeros.

We put an example dataset (part of the AIOps18 dataset due to the space limit) in the ``data'' director.

Tree

.
├── dataload.py
├── detect.py
├── extract_feat.py
├── model
│   ├── PixelCNN.py
│   ├── detection.py
│   ├── diffusion.py
│   ├── network.py
│   ├── prediction.py
│   └── util.py
├── predict.py
├── requirements.txt
└── util.py

Environment

We support python3.x $\geq$ 3.7. The environment can be built by: $ pip install -r requirements.txt

Running

For the prediction phase, run: $ cd src && python predict.py

For the detection phase, you should first extract features and then run: $ cd src && python detect.py