Pytorch implementation of Resnet for time-series prediction and use in Numerai tournament.
Based on: https://www.kaggle.com/a763337092/pytorch-resnet-starter-training
Several users on Numerai forums have found success using architectures with residual connections. This implementation isbased off an architecture used in Janestreet Market Prediction competition hosted on Kaggle. While not the perfect implementation of a ResNet, this implementation works well with multivariate time series data.
Data is pulled from Numerai's API for python - not hosted in this repository (dataset is too large to host on github).
ToDo:
- Feature-Target neutralization