A Pytorch implementation of DeepLog's log key anomaly detection model.
If you are confusing about how to extract log key (i.e. log template), I recommend using Drain which is proposed in this paper. As far as I know, it is the most effective log parsing method. By the way, there is a toolkit and benchmarks for automated log parsing in this repository.
- python>=3.6
- pytorch==1.4
- tensorboard==2.0.2
The dataset can be downloaded HERE.
The prefix of the website for hdfs logs is changed to http://people.iiis.tsinghua.edu.cn/~weixu/, you can download the original data from raw log and 200nodes.
Run the following code in terminal, then navigate to https://localhost:6006.
tensorboard --logdir=log
Min Du, Feifei Li, Guineng Zheng, Vivek Srikumar. "Deeplog: Anomaly detection and diagnosis from system logs through deep learning." ACM SIGSAC Conference on Computer and Communications Security(CCS), 2017.
If you have any questions, please open an issue.
Welcome to pull requests to implement the rest of this paper!