Skip to content

mitdbg/logos

Repository files navigation

LOGos

Utilizing system logs to perform causal analysis. You can access the documentation here.

Please begin by installing the Python packages required for this project by running pip install -r requirements.txt.

OpenAI integration

In order to use the LLM-powered capabilites of LOGos, please add a .env file to the root of this repo and define OPENAI_API_KEY appropriately.

Trying out LOGos

For an introduction to our Python-based interface, you can turn to our demo notebook at demo/demo.ipynb.

We also offer a simple UI built using Streamlit. You can launch it by running demo/run_ui_demo.sh and following the resulting URL.

Reproducing our evaluation

To reproduce the evaluation from our VLDB paper, please follow the following steps:

  1. Follow the instructions in dataset_files/README.md to gain access to our datasets.
  2. Within evaluation/, you will find directories based on each experiment presented in our paper. Based on the experiment you would like to reproduce, switch into the appropriate directory and run the reproduce.sh script (you may need to edit file permissions to make it executable). This will run the experiment and plot the results.
  3. Find the resulting plots in evaluation/repro_plots/. The raw data for each plot will be saved in evaluation/repro_plots_data/.

About

Human-in-the-Loop Causal Analysis of Log Files

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published