Checklist ONIA 2026 National Stage · Task 3
The Ghost in the Machine
Romanian title: Fantoma din Mașinărie
Predict imminent server failures and an anomaly score from noisy monitoring metrics, alert titles and semi-structured system logs.
The task
A site reliability engineer at a company running a 200-server cluster inherits chaotic monitoring data: partially corrupted numeric metrics, inconsistent category labels, free-form alert titles written by different engineers and multi-line semi-structured system logs.
Task A asks whether each server will fail within the next four hours (labels are missing for about 40% of training rows); task B asks for an anomaly score between 0 and 100. The test file covers a new monitoring period.
Abridged and translated by SOTA from the official Romanian materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
train.csv(6,000 rows: id, timestamp,server_id, region,server_type,cpu_usage,memory_usage,disk_io,network_latency,error_rate, temperature,alert_title,system_log, failure,anomaly_score) andtest.csv(1,500 rows, without failure andanomaly_score). Archive of 830,653 bytes.- You submit
- CSV with id, subtaskID (1 = failure, 2 =
anomaly_score) and answer; two rows per test id; failure answers must be 0 or 1. - Scoring
- Task A (50): F1 for class 1; 0 if F1 ≤ 0.3, 50 × (F1 − 0.3) / 0.3 for 0.3 < F1 < 0.6, 50 at F1 ≥ 0.6. Task B (50): MAE; 50 at MAE ≤ 8, 50 × (13 − MAE) / 5 for 8 < MAE < 13, 0 at MAE ≥ 13.
- Rules
- On-site at the National University of Science and Technology POLITEHNICA Bucharest
- No internet access; offline resources only (NLTK, spaCy
en_core_web_sm, scikit-learn, Gensim glove-wiki-gigaword-50, fastText without pretrained models, distilbert-base-uncased) - Downloading models or data is forbidden; allowed documentation listed in the round description
- Problem statements unlocked with a password handed out in the room
- Format
- ONIA 2026 national stage, 18 April 2026; same problem for both grade sections.