Checklist HAIO 2025 Summer National Final · Task 5
Classifier Clone
English title: Klasszifikáló Klón
Train a non-neural machine-learning model that reproduces the predictions of a given pretrained neural classifier as closely as possible.
The task
A pretrained neural network (TorchScript, secret_model.pt) was trained for a classification task. The contestant receives the network, its training data (X_train, y_train) and a small subset of the test data with labels, and must build a model that is not a neural network and that imitates the network's outputs on the full test set.
Scoring uses the seed set in the first cell (42), and every notebook is run only once during evaluation; setting a random state for the models is recommended. A separate train–test split file may be used only at the end, for the final evaluation, and not for training or development.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
In English
Some of this task's files were published only in Hungarian. SOTA translated that file into English on 16 September 2026.
Read the task description (one-pager) in English
Classifier Clone
English translation by SOTA – AI Community of the Hungarian original. Licensed CC BY-NC-SA 4.0, like the original. Organisers who would like this translation removed can email [email protected].
Magyar MI Diákolimpia (Hungarian AI Olympiad)
Classifier Clone (Klasszifikáló Klón)
Task description
24 May 2025
1. Classifier Clone
You are given a pretrained neural network, which we trained for a classification task on a specific training dataset. To solve the task, you have at your disposal:
- the trained neural network;
- the training dataset (
X_train) and labels (y_train); - a small subset of the test dataset (
X_train_small) and labels (y_train_small).
The goal is to build a machine learning model that is not based on a neural network. The job of this model is to approximate the output of the trained neural network as closely as possible on the full test dataset, that is, to try to imitate its behaviour.
2. Scoring
The final model will be evaluated on the full test dataset. Scoring is based on the accuracy metric, according to the following scale (the right end of each interval is open):
- 0 - 0.5: 0 points
- 0.5 - 0.62: 20 points
- 0.62 - 0.66: 30 points
- 0.66 - 0.68: 40 points
- 0.68 - 0.70: 50 points
- 0.70 - 0.71: 70 points
- 0.71 - 0.728: 80 points
- 0.728 - 0.80: 90 points
- 0.80 - 1: 100 points
Maximum score: 100 points
During scoring, the seed set in the first cell will be used. In addition, it is recommended to set a random state for the models as well, since this can affect the results they achieve. During evaluation, all code will be run only once.
3. Technical information
To solve the task, the attached .ipynb notebook file, the weights of the secret model and the dataset are sufficient. For evaluation, a file named train-test.pt is also provided, but you may use it only at the end of your solution, to evaluate the model. It is forbidden to use it during training or development. To help you, several useful libraries and guides can be found at the beginning of the notebook.
After solving the task, regardless of whether you worked in Google Colab or locally, you must upload the .ipynb file to the CMS system for the corresponding task. No other file needs to be uploaded.
4. Useful resources
Translated by SOTA. The Hungarian original is the official version and wins wherever the two differ. Original by the Hungarian AI Olympiad (ELTE Faculty of Informatics), licensed CC BY-NC-SA 4.0; this is a translation of the task one-pager, and the official English notebook of the task is linked on this page. This translation is shared under CC BY-NC-SA 4.0, the licence of the original. 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
secret_model.pt(TorchScript network),data.pt(loaded as X_train, X_test_small,y_train,y_test_small) and the evaluation split file, all downloaded with gdown; the three files are also in the repository folder adatok/klasszifikalo-klon.- You submit
- The completed .ipynb notebook uploaded to the CMS.
- Scoring
- Accuracy on the full test set, with right-open intervals: [0, 0.5) → 0 points; [0.5, 0.62) → 20; [0.62, 0.66) → 30; [0.66, 0.68) → 40; [0.68, 0.70) → 50; [0.70, 0.71) → 70; [0.71, 0.728) → 80; [0.728, 0.80) → 90; [0.80, 1] → 100.
- Rules
- The final model must not be based on a neural network.
- The evaluation split file may be used only for the final evaluation.
- Work only on the provided lab machines (Windows); own laptops are not allowed. Google Colab is recommended; a local Python environment is also provided.
- At most one NVIDIA T4 GPU; stronger GPUs (e.g. A100, V100) lead to disqualification.
- General internet access, but code-completion and LLM services (e.g. GitHub Copilot, ChatGPT, Claude) are forbidden; the free tier of Gemini 2.5 Flash integrated in the Google ecosystem is the only exception.
- Publicly available documentation, articles and books may be used during the practical part; audio-visual material is forbidden; communication is forbidden.
- Mandatory screen recording with OBS Studio for the whole contest.
- Format
- Summer National Final (Nyári Országos Döntő), 24 May 2025, on site at ELTE, Budapest. Second (practical) part: two hours of programming after a 15-minute briefing; four 100-point tasks (CV, ML, NLP, RL). Solutions (.ipynb) are uploaded to the CMS.