Checklist NEOAI 2025 Day 1 · Task 1
Tricy Table Data
Improve a fixed LightGBM regressor on incomplete tabular data purely through data preprocessing, target transformation and sample weights.
The task
The legend ("Slava's glass, Part 2") frames a standard tabular regression problem in which parts of the data are missing or corrupted, and only gradient boosting may be used. The training set has 9 numeric features, 3 datetime features and the target; the test set has the same features and an id.
The final submission must be produced by the supplied function clf_train, which trains a LightGBM model with fixed parameters (MAE objective, 500 boosting rounds) on the contestant's preprocessed features, target weights and optional target transformation, and writes the submission. Only the inputs to this function may be engineered.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
train_tables.csv(9 numeric features, 3 datetime features, target),test_tables.csv(same features, id),sample_submission.csv.- You submit
- CSV with columns id, target, generated by
clf_train. - Scoring
- SCORE = (Σ(
true_i−predict_i)² / n)^(1/4), i.e. the square root of the RMSE (lower is better; the notebook calls it "squared RMSE"). - Rules
- The code of
clf_trainmust not be changed, and only submissions produced by it may be used. - General NEOAI 2025 rules apply: by default no pretrained models or additional data unless the statement allows them; no external human help; on-site participants get one NVIDIA V100, an LLM chat (GPT-4o or similar) and a whitelist of websites.
- The code of
- Format
- NEOAI 2025 main round, Task 1, Day 1 (7 May 2025, 07:00–13:00 UTC). Individual; 20 submissions per day.