We trained a large ensemble of deep learning frameworks where FFNNs, CNNs and Transformer models are included to forecast the return rates of 3773 investment assets in the market. We also utilized this model in the Ubiquant Market Prediction competition on Kaggle to learn 300 anonymized features and achieved 41/2893 in the public leaderboard ranking (top 2%).
This repository contains Tensorflow implementation and architecture illustration of our model.
This work is also a group project for CSC413- Deep Learning and Neural Networks taught by Prof. Jimmy Ba (https://jimmylba.github.io/) and Prof. Bo Wang (https://www.wanglab.ml/).
Details about the challenge can be found here: https://www.kaggle.com/competitions/ubiquant-market-prediction/overview