AIAdventures / keras-for-numerai

Participation in the Numerai competition with Keras

Keras for numerai

See blog post

Contains starter code for participation in the Numerai competition using deep learning framework, Keras

1. Configuration

• Define your Neural Network Architeture in train.py
• Download data from Numerai
• Remove t_id column in numerai_tournament_data remove feature column in numerai_training_data
• Put data in /data

Dependencies

• pip install requirements.txt

Set variables training

• validation_split=0.3, Sets aside 30% of training data for testing
• shuffle=True, Set to True if you want to shuffle data before each epoch
• nb_epoch=50, Number of epochs you want your neural net to train
• batch_size=5000 Number of batches you would like to use during training

2. Usage

Train the network python train.py

Create predictions python predict.py predictions are in /predictions/predictions.csv

3. Performance details

Varies by architecture used & Numerai data released each week

4. Links

Wide and Deep Neural Net

Wired Article on Numerai