Keras-Tensorflow boilerplate starter for the Numerai competition.
I wanted to use Keras for the Numerai tournament and couldn't find a basic example so I made this one.
If you are new to the competition and/or want a better explanation of the code see this companion post:
Practical Keras: Simple regression for the Numerai Tournament
There's a separate script to calculate Validation Correlation (val_corr.py) which should match the one generated by Numerai.
And some barebone metrics (metrics.py) contributed by parmarsuraj99 ( see also this forum post )
Out of the box this NN/ Model is not very good, you need to tune parameters, add your preferred validation, tweak the model, etc, etc see the following section...
Open an issue or pm on the Numerai chat if you want to add your contributions.
Ideally each new contribution is a standalone script with a new feature, some missing features :
- Save the model ( as to not retrain )
- Cross validation ( within the dataset and with val1 val2 )
- Payout scores and metrics (see Validation)
- Parameter tuning.
- Colab/jupyter notebook.
- Your suggestion ?
It would really be cool to have a community model to go against example_predictions, just saying...
Toss a 💰to your 🧙er !
NMR/ETH: 0xD2468b070e27138525ae347410A5cFFcEDc3d0E7