Numerai Team Overview

The Numerai team is a diverse group of individuals from various backgrounds, including mathematicians, data scientists, and engineers. They bring together a wide range of skills and expertise to tackle complex problems in machine learning and artificial intelligence.

Numerai's co-founders, Adam Degtear and Alex Pack, have a strong background in finance and technology, which has helped shape the company's mission and approach. They aim to create a more transparent and fair financial system through the use of machine learning and data science.

The team is headquartered in San Francisco, but they have a global presence with team members working remotely from all over the world. This allows them to tap into a diverse pool of talent and perspectives, which is essential for tackling complex problems in AI and machine learning.

You might enjoy: Numerai

The numerai team has been making waves in the data science and machine learning communities. They're known for hosting a unique tournament-style competition where participants can test their skills and compete for cash prizes.

numerai has a strong focus on community engagement, with a dedicated team of moderators and developers working to improve the platform and create a supportive environment for users. They're also committed to transparency, with a clear and open approach to how they handle data and model submissions.

numerai's tournament-style competition has been running for several years, with thousands of participants from all over the world taking part. The platform has seen significant growth and adoption since its inception.

The Numerai team has developed Benchmark Models that are a set of standard models built by the team, which are then given out every round so that anyone can easily submit them and stake on them if they want.

These Benchmark Models are an easy way to compare your model to the current state-of-the-art. They're a great reference point to see how your model stacks up.

All predictions made by Numerai are done using a Walk-Forward framework, which means models are trained only on data available prior to the date of the prediction being made.

This framework is implemented by splitting the data into chunks of 156 eras, and then training a model on all eras up to the first era of the chunk minus a certain number of purge eras.

For 20D targets, the number of purge eras is always 8, and for 60D targets, it's always 16.

The walk-forward validation windows are quite complex, but essentially, a model is trained on a certain range of eras, then predicts on a subsequent range of eras, and so on.

The Standard Large LGBM params model is a simple 50/50 blend of two other models, V5_LGBM_TEAGER2B20 and V5_LGBM_CYRUSD20.