Spaces:
Runtime error
Welcome to the Wyze Rule Recommendation Challenge!
We're excited to host this machine learning competition focused on developing innovative rule recommendation systems for smart homes. As outlined in the challenge overview, the goal is to leverage Wyze's rich smart home dataset to train models that can predict customized automation rules tailored to each user's unique devices and behaviors.
To get started, we have put together a sample baseline model in this GitHub repo: https://github.com/yh-yao/FedRule.git
This implements the GraphRule algorithm from the paper: FedRule: Federated Rule Recommendation System with Graph Neural Networks. The code demonstrates how to:
- Load and preprocess the Wyze dataset
- Construct user-rule and user-device graphs
- Apply graph neural network techniques
- Train a centralized model on the graph embeddings
- Make rule recommendations for new users
This is intended as a simple starter code example for the competition. Please refer to the FedRule paper for more details.
There are many other innovative modeling techniques that could be applied here. Feel free to leverage any machine learning approaches that can effectively generate personalized rule recommendations.
We are looking forward to seeing the creative solutions you all come up with! Please use this discussion thread for any questions, ideas, or updates on your progress. Best of luck!
Hi @mmkamani7 ,
Thank you for this very interesting competition. I wonder why so few people are participating ... ?
For curiosity, do you know how well your baseline performs on the public leaderboard ? As for me, I think I will stop improving now. I'm happy to have crossed 0.45x !
Edit : for the private phase, will we need to submit other files or do I just need to wait now ?
Thanks for your participation. The lower level of participation may be attributed to limited visibility and promotion channels for the competition that we have access to. We're glad you find this challenge intriguing, and we're looking forward to witnessing your innovative approach.
I can say that your score has already passed our baseline, which was around 0.25. I hope you can continue your effort to pass the 0.5 mark, which is huge due to the complexity of the task.
As for the private phase, participants will be provided with a fresh test set. When this phase begins, you will need to submit distinct files specifically tailored for this new test set. We appreciate your engagement and look forward to your continued participation.
Thank you for your answers.
Regarding the 0.5 mark, I have already passed it in my cross-validation scheme, which has been really correlated to the public LB so far.
However, as I said, I'm happy with the 0.45+ LB score for now. To be transparent, I still have some ideas that would improve the score, but I don't want to invest more time on this project - interesting as it is ! To give you an idea, I already spent about 15-20 hours building my solution, so it wouldn't been cost-effective to continue at this point, considering the prizes and the low participation rate.
However, if you want to push the limit, I would be glad to work with you as an independent. Feel free to contact me here or via Linkedin/Kaggle 🙂