Numerai Model Uploads
Model Uploads is a new feature that makes it easy to automate your daily submissions for the Numerai Data Science Tournament.
Deploy your model straight from Google Colab and let Numerai handle generating live predictions for you every day.
Free, zero-setup automation — so you can focus on building machine learning models to predict the stock market.
System Architecture
At the core of Numerai is our trading pipeline that runs every day.
It starts with Numerai transforming raw financial data into clean and regularized live features for the tournament.
Next, users download these live features, feed them into their trained models for inference to produce live predictions, and submit those predictions back to Numerai.
Finally, Numerai combines all predictions to create the Stake-Weighted Meta Model Signal, and feeds it to the portfolio optimizer to generate trades.
Infrastructure Tax
The unique thing about Numerai’s system architecture is that user models live outside of our infrastructure. This design ensures that users have the maximum flexibility when building their machine learning models.
But what this means practically is that every user must set up and maintain their own infrastructure to host their model. To get a sense of the complexity of a typical setup, check out the official Numerai Compute framework or this community-contributed DIY tutorial.
The need to set up and maintain this infrastructure is a huge tax on the time and energy for our existing users, and a high barrier to entry for new users looking to get started.
This problem became especially clear when we transitioned from weekly to daily trading at the end of 2022. Many existing users with manual submission workflows had to scramble to get automated. Many new users simply gave up and quit.
Model Uploads
How can we abstract away the complexities of setting up and maintaining this infrastructure without sacrificing flexibility?
The solution is to simply let users deploy models onto infrastructure set up and maintained by Numerai.
Here is how it works. First, pickle your model along with any code required to run it. See a full working example of this on Google Colab.
Then, just upload the pickle file and Numerai will run it to generate live predictions every day.
Moving forward, use this simple and intuitive dashboard to track your submissions and view validation diagnostics for each uploaded model.
That’s it! Did we mention this is free?
Get Automated Today!
Whether you are just getting started or looking to simplify your workflow, it has never been easier to get automated for Numerai.
Follow our brand new onboarding tutorial to build and upload your first model and earn 0.1 NMR in stake credits!
