You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Lograph GNN Checkpoints

This repository contains trained checkpoints for a reproduced supervised Graph Neural Network implementation of Lograph, following the method proposed in Anomaly Detection on Interleaved Log Data With Semantic Association Mining on Log-Entity Graph by Guojun Chu et al.: https://github.com/GeorgeChu2019/Lograph. The repository includes models trained on the BGL, HDFS, OpenStack, and Thunderbird datasets, each evaluated with three random seeds: 42, 123, and 2026.

Repository Contents

For the BGL and HDFS datasets, both grouped and ungrouped configurations are included. In the grouped setting, grouping is performed by node_id for BGL and by block_id for HDFS. For the OpenStack and Thunderbird datasets, only ungrouped configurations are provided.

To study the effect of preprocessing, the models were trained with two parsing variants: the manual parsing technique published by the authors of the original paper and a Drain-based parsing variant added in this work for comparison. Both variants are available for all datasets.

For BGL, HDFS, and OpenStack, checkpoints are provided for three dropout settings: 0.0, 0.1, and 0.2. For Thunderbird, only the best-performing configuration for each parsing variant is uploaded: dropout 0.2 for the manual parsing variant and dropout 0.2 for the Drain-based variant.

Datasets

The uploaded checkpoints cover the following datasets:

BGL
HDFS
OpenStack
Thunderbird

Experimental Variants

The repository includes checkpoints across the following dimensions:

Three random seeds: 42, 123, and 2026
Two parsing variants: manual and drain
Grouped and ungrouped settings where applicable
Multiple dropout settings where available

Notes

These checkpoints are provided as part of a Master's thesis reproduction and extension study of the original Lograph method. In addition to reproducing the original setup, this repository includes comparative experiments with a Drain-based parsing pipeline.

Citation

If you use these checkpoints, please cite the original Lograph paper and reference the corresponding thesis work.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support