log-classifier

Classify production error and log lines into categories for automatic triage and routing (e.g. to the right team or runbook).

Categories: auth · timeout · config · bug · dependency · rate-limit

Model: TF-IDF + LogisticRegression on message text. Trained on a curated dataset plus optional Loghub logs (Linux, Apache) with rule-based labels.

---

Setup

python -m venv venv
venv\Scripts\activate   # Windows
# source venv/bin/activate   # Linux/macOS
pip install -r requirements.txt

Data

• Curated: data/labeled_logs.csv (columns: message, category). Included in the repo.

• Optional – Loghub: Download e.g. Linux (2.25 MiB) or Apache (4.9 MiB), extract, and put the log file(s) under data/loghub/Linux/ or data/loghub/Apache/. Then run:

  python scripts/process_loghub.py

  This creates data/loghub_labeled.csv. The trainer merges it with the curated CSV automatically.

Train

python -m src.train

Uses data/labeled_logs.csv (and data/loghub_labeled.csv if present). Prints accuracy, F1, per-class metrics, and confusion matrix. Saves the pipeline to models/log_classifier.joblib.

Classify

python -m src.classify "Connection timeout after 30s"

Outputs suggested category and confidence. Requires a trained model (python -m src.train first).

Results and analysis

Open notebooks/results_analysis.ipynb and run all cells to see metrics, confusion matrix, and example correct/wrong predictions with short notes on why the model fails in some cases.