ML Academic Project applied to the Kaggle Challenge : Jane Street Real-Time Market Data Forecasting 2024.
The goal is to compare two ML models on the same prediction task, which is to predict one of the responders up to 6 months in the future. The models tested are Temporal Fusion Transformer (TFT) and Variational Autoencoders (VAE).
To make the code work, you first need to download the data frome Kaggle here and put the parquets inside a raw_data/train_parquet folder. Then you'll be able to launch the code/preprocessing.py to create preprocessed training and validation datasets that we will use in the notebook.