Checklist OFIA 2025 Stage 1 (first online selection round) · Task 1
Mushroom Mission on ExoMycelia-7
French title: Mission Champignons sur ExoMycélia-7
Predict the probability that each test mushroom is toxic from tabular measurements and a photograph of the mushroom.
The task
After a long journey through space, Raton (a raccoon) and Renard (a fox) land on the planet ExoMycélia-7 and explore a forest full of strange, luminous mushrooms, some delicious and some extremely toxic. Raton's 'Champidex' holds information on mushrooms found on other planets and indicates whether each is edible or poisonous, but it does not yet know the mushrooms of this planet. The two friends photographed and measured the mushrooms they picked, and the contestant must train a model that predicts the probability that each unknown mushroom is toxic.
The training set is the Champidex, with 10,000 mushrooms; the test set contains the 5,000 mushrooms picked by Raton and Renard. For each mushroom a photograph is provided together with measured characteristics: position (x, y), weight in grams, odour, texture, soil type, porosity (0 to 1, where 1 means very porous), pH of the juice, pH of the soil, soil temperature (°C), presence of insects (yes/no), effect on touch (nothing, stings, burns slightly, …) and whether the mushroom looks delicious according to Renard (yes/no). The images may contain information useful for the prediction in addition to the columns.
Using the training data (CSV files and images) and the test files, the contestant writes Python 3 code that computes, for each test mushroom, the probability that it is toxic; the resulting CSV is scored by Kaggle. The organisers reserved the right to change the data, in particular the test file, up to three days before the end of the round.
Abridged and translated by SOTA from the official French materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
X_train.csv(id, x, y, poids, odeur, texture,type_sol, porosite,ph_du_jus,ph_du_sol,temperature_du_sol,presence_insecte,effet_au_toucher,a_l_air_delicieux_selon_renard);y_train.csv(id,est_toxique= 1 if toxic, else 0);images/train/champignon_i.png;X_test.csv;images/test/champignon_j.png;exemple_de_soumission.csv;generateur_exemple_de_soumission.py;exemple_lecture_image.py.- You submit
- CSV with header id,
probabilite_toxiquegiving a toxicity probability between 0 and 1 for every test id (ids are numbered from 1). - Scoring
- ROC AUC between the predicted probabilities and the observed toxicity, converted to a final score = (ROC_AUC − 0.5) × 200. During the live round the qualification threshold for Stage 2 was set at 40 points on the private leaderboard, within a limit of 100 qualified pupils (announcement of 22 March 2025).
- Rules
- Python 3; suggested libraries: scikit-learn, PyTorch, TensorFlow, NumPy, pandas, Matplotlib, Seaborn.
- AI assistants may be used for reasoning and coding, but candidates must be able to explain their ideas and code.
- Code must run without third-party services: calls to external APIs (e.g. GPT-4, Hugging Face) are forbidden; libraries and pre-trained models may be included in the code.
- Up to 100 submissions per day; 2 submissions selected for final evaluation.
- Individual participation only; discussing the task with others during the round is forbidden.
- The code behind the final submissions (including feature computation) had to be uploaded, together with a school certificate.
- Open to French nationals or pupils studying in France who are under 20 on 2 August 2025 and not in higher education.
- Format
- OFIA 2025 Stage 1, online on Kaggle from 7 Mar 2025 (10:00) to 30 Mar 2025 (23:59). The Stage 1 score counted for one third, and the Stage 2 score for two thirds, of the evaluation used to select finalists. The public archive competition replays the round for practice.