Refactor private repositories to ml4logs so we have perfectly reproducible experiments.

LogAnalysisTeam / ml4logs

Machine Learning methods for log file processing

MIT License

0 stars 3 forks source link

Refactor private repositories to ml4logs so we have perfectly reproducible experiments. #1

Closed dhonza closed 3 years ago

dhonza commented 4 years ago

We need to get easily reproducible results on both HDFS and BGL datasets:

[x] Drain-based LAD methods from the loglizer suite.
[x] Supervised fastText embedding LAD,
[x] Unsupervised fastText embedding LAD
[x] Check that the dataset splits use fixed random seeds.
[x] Fix data/embedding download scripts.
[x] Fix scripts to run all benchmarks automatically.
[x] Fix documentation.

dhonza commented 3 years ago

The whole project was rewritten. It is now based on pipelines and all experiments are configured by JSON files. Data splits are reproducible and most intermediate data representations are stored in the filesystem.