Numerai Model Survey Results

Hi all! I recently posted a short survey to find out what types of models people are using and since there aren’t too many more results coming in I have decided to publish the results.

I will now rank the types of models in order of popularity*;

  1. Gradient Boosting (XGBoost, Catboost, etc.) - 69.6%
  2. Keras Neural Network - 52.2%
  3. Random Forest - 17.4%
  4. AutoML - 4.3%
  5. Ridge - 4.3%
  6. ElasticNet - 4.3%
  7. Naive-Bayes - 4.3%
  8. PyTorch NN - 4.3%
  9. Jax/Flax NN - 4.3%
  10. Genetic - 4.3%
  11. Linear - 4.3%

*Percentages do not add up to 100% due to multi-model/ensembling

It is also interesting to note that 47.8% of people stated that they used an ensembled model.

The final question was a bit more varied, with only a handful of people training on Nomi since its release. Feature neutralisation was also hit and miss with some users reporting that they had stopped neutralising in order to improve results.

I hope this was useful/interesting to someone other than me!