Practice training neural networks in Theano/Keras for Numerai submissions. (https://numer.ai/)
Uses feed-forward neural network with 1 hidden layer, sigmoid activation function to predict Numerai scores.
Number of hidden layer nodes and regularization parameter selected using manual k-fold cross validation.
Requirements:
sudo pip install keras
Command-line options:
-n: [number of possible nodes in hidden layer]
-r: [possible values of lambda]
-k (optional): number of folds in cross validation
Example:
python main.py -n 50 75 100 150 -r .01 .015 .02 .025 -k 10