Checklist NTO AI 2026 Stage 1 (qualifying), engineering tour · Task 1
Classification of Space Objects
Russian title: Классификация космических объектов
Classify synthetic celestial objects into six classes, including a rare exoplanet-candidate class, from photometric, kinematic and spectral features.
The task
Telescopes collect data on millions of celestial bodies, too many to analyse by hand. Participants build a model that determines the class of a celestial object from its numerical characteristics: main_sequence_star, red_giant, white_dwarf, galaxy, quasar or exoplanet_candidate (rare).
The data are synthetic, with noise, missing values, correlations and class imbalance. There are 25–35 numerical and categorical features, including u/g/r/i/z magnitudes and colour indices, parallax, proper motions, radial velocity, per-band signal-to-noise, observation count, cloud factor, background noise, H-alpha, [O III] and Na D line strengths, extinction and a categorical quality_flag.
Abridged and translated by SOTA from the official Russian materials. The official statement has the exact rules, and it wins wherever this summary differs.
In English
This task was published in Russian. SOTA translated its 4 files into English on 16 September 2026. Only the words changed in the notebooks: markdown, code comments, messages and printed output. The code, file names and paths are the original's, so a translated notebook runs with the original data.
- Task statement Russian original of Task statement
- Baseline notebook Russian original of Baseline notebook
- Official solution (from the task collection) Russian original of Official solution (from the task collection)
- Solution notebook Russian original of Solution notebook
Read the task statement in English
Classification of Space Objects
English translation by SOTA – AI Community of the Russian original. Organisers who would like this translation removed can email [email protected].
2.3. Engineering tour
Task 1 (100 points)
Classification of space objects
Topics: artificial intelligence, machine learning, data processing, classification.
Statement
Background
Modern telescopes at observatories collect terabytes of data every day about millions of celestial bodies: their coordinates, brightness and spectra. Among them are ordinary stars as well as quasars, galaxies and even exoplanet candidates, which are especially interesting and important objects for science.
However, the volume of data is growing so fast that astronomers cannot analyse it by hand. To avoid missing valuable finds, an automated approach is needed that can classify objects quickly and reliably by their characteristics.
Your model will help to speed up scientific discoveries and make life easier for astronomers.
What needs to be done?
Create a model that determines the class of a celestial object from its numerical characteristics.
Classes to predict:
- main_sequence_star — main-sequence star;
- red_giant — red giant;
- white_dwarf — white dwarf;
- galaxy — galaxy;
- quasar — quasar;
- exoplanet_candidate — exoplanet candidate (a rare class).
Data
Download the data: https://disk.360.yandex.ru/d/mXy_XHfYrZtZMA.
The following files are available to participants:
- train.csv — the training set with labels:
features (25–35 in total): numerical and categorical; some of them are:
- photometry: u_mag, g_mag, r_mag, i_mag, z_mag;
- indices (differences of magnitudes): u_g, g_r, r_i, i_z;
- geometry / motion: parallax, pm_ra, pm_dec, radial_velocity;
- signal / noise: snr_u, snr_g, snr_r, snr_i, snr_z, obs_count, cloud_factor, background_noise;
- "spectral hints": h_alpha_strength, oIII_strength, na_d_strength;
- auxiliary: extinction, quality_flag (categorical);
- test.csv — the test set (without the target label);
- sample_submission.csv — an example file with predictions.
The data are synthetic: they contain noise, missing values, correlations and class imbalance.
Prediction format
The file submission.csv must contain:
-
the column object_id;
-
one column for each class, with probability 0 or 1.
Example:
object_id,main_sequence_star,red_giant,white_dwarf,galaxy,quasar,exoplanet_candidate
123456,0,0,0,1,0,0
... -
First column: object_id.
-
Then: the probabilities of belonging to each class (0 or 1).
-
The number of rows equals the number of objects in test.csv.
Metric
This is a multiclass classification task. The quality of a solution is evaluated with the macro F1 score metric, which computes the F1 score for each class and averages the result.
Baseline solution example
The file baseline.ipynb contains a baseline solution example (a baseline) that can serve as a starting point for developing your own solutions.
Baseline solution example: https://disk.360.yandex.ru/d/CdN8a5oXxUgN-g.
The collection's solution for this task is in the separate solution file.
Translated by SOTA. The Russian original is the official version and wins wherever the two differ. The statement is the Stage 1 engineering-tour task (pages 88–90 of the official 2025/26 task collection); the data are on the organisers' Yandex Disk. Both notebooks train and cross-validate a CatBoost model on train.csv and save it, but neither writes submission.csv; the solution notebook is the author's macro F1 ≈ 0.95 version (baseline_95.ipynb). If you organise this olympiad and would like the translation removed, email [email protected] and we will take it down.
At a glance
- You get
train.csv(features and label),test.csv,sample_submission.csv;baseline.ipynb.- You submit
submission.csvwith columnobject_idand one 0/1 column per class (main_sequence_star,red_giant,white_dwarf, galaxy, quasar,exoplanet_candidate); one row per test object.- Scoring
- Macro-averaged F1 score. The collection reports about 0.75 for the CatBoost baseline and 0.95 for the author's solution.
- Rules
- Individual task.
- Format
- Stage 1, engineering tour (100 points); the mathematics, informatics and engineering tours are each worth 100 points and 50 in total are needed to advance.