Hi everyone

Following the recent discussion and helpful questions and comments from the community, we’d like to share an update and clarification with everyone:

The original question was:

“Given there are 15 targets, can we train 15 specialised models (one per target), or should we train a single model that predicts all 15 targets? We want to be sure we won’t be disqualified during evaluation.”

Updated guidance:

To balance real-world mobile deployment with modelling flexibility, we are allowing solutions that use one or a small number of models (typically up to 3 or 4), where each model predicts a group of related rice quality metrics.

👉 In other words:

• ❌ 15 separate models (one per target) is not recommended
• ✅ 1 multi-output model recommended or
• ✅ 3–4 sub-models, each covering a coherent set of rice quality measures

Approaches involving a large number of independent models (e.g. one model per target) are not aligned with the spirit or intended real-world use case.

Thank you to everyone who raised this question - it helped us refine the guidance in a way that’s both practical and fair. We aim to host challenges that are both technically interesting and practically usable for clients, and we appreciate the community’s engagement in helping us strike that balance.

Good luck, and happy modelling 🚀