numerai_analyser

This package automates the weekly competitions on numer.ai.

Getting Started

python run.py

Summary

The package uses numerapi to download the latest competition data, then tests a range of applicable models and features before automatically uploading the predictions from the best model.

Authors

• Glen Moutrie

Plan:

1. create a data loader that creates a NumeraiData class -- DONE
2. Define a NumeraiData Class with features, x data and y data -- DONE
3. finish the model tester that returns the model results, estimating in parallel using multiprocessing
4. automate with numerapi and other such tools -- DONE
5. Better predictive models, look at alternaitves, betters model specifications
6. Automatic feature selection --DONE
7. Feature engineering (look at clustering etc)
   1. Clustering -- DONE
   2. Add principal components
   3. Add predictions from other models
8. Fix the cross terms issue by removing white space in names
9. Potentially try a deep learning approach...
10. Try an ensemble approach accross different epochs
11. Improve unit tests and logging