Add dataset card
Browse files
README.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-4.0
|
| 3 |
+
pretty_name: "IOAI 2026 – IOAI Field"
|
| 4 |
+
tags:
|
| 5 |
+
- ioai
|
| 6 |
+
- olympiad
|
| 7 |
+
- regression
|
| 8 |
+
viewer: false
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# IOAI 2026 — IOAI Field
|
| 12 |
+
|
| 13 |
+
Official contest data for **IOAI Field**, task 6 (Day 2) of the [IOAI 2026](https://ioai-official.org) Individual Contest, held in Astana, Kazakhstan.
|
| 14 |
+
|
| 15 |
+
The IOAI logo is modelled as a spatial function F(x, y, W) over the unit plane. Contestants train a small network **from scratch** (no pretrained weights, ≤ 20,260 parameters) to reproduce the field; the grader overlays hidden leaderboard configurations onto the training config and retrains the submitted notebook.
|
| 16 |
+
|
| 17 |
+
The task statement, translations, baseline and grader live in the [IOAI-2026 GitHub repository](https://github.com/IOAI-official/IOAI-2026/tree/main/Individual-Contest/6_IOAI_Field). This repository holds **data only** and is **not published**.
|
| 18 |
+
|
| 19 |
+
## Subsets
|
| 20 |
+
|
| 21 |
+
| Subset | Path | Contestants saw it | Role |
|
| 22 |
+
|---|---|---|---|
|
| 23 |
+
| Public – train | `public/train_config/`, `public/train.npz`, `public/test.npz`, `public/meta.json`, `public/preview.png` | yes | the visible field configuration and samples |
|
| 24 |
+
| Private – held-out test | `private/test_private.npz` | no | hidden evaluation queries |
|
| 25 |
+
| Private – Leaderboard A | `private/leaderboard_a_config/` | no | hidden field config for the **live leaderboard** |
|
| 26 |
+
| Private – Leaderboard B | `private/leaderboard_b_config/` | no | hidden field config for the **final ranking** |
|
| 27 |
+
|
| 28 |
+
> ⚠️ **The `private/` subset contains the hidden evaluation configurations — effectively the answers.** This dataset is kept private so the task remains reusable.
|
| 29 |
+
|
| 30 |
+
## Credits
|
| 31 |
+
|
| 32 |
+
Proposed by **Evgenii Tsymbalov**; developed by **Evgenii Tsymbalov**, **Ekaterina Fadeeva**, **Daniil Kazantsev**, **Maiya Goloburda** and **Magauiya Zhussip**. Published by the IOAI 2026 Scientific Committee under CC BY 4.0.
|