In this work I implemented, using pytorch, the models (DeepLOB-Seq2Seq and DeepLOB-Attention) proposed in Multi-Horizon Forecasting for Limit Order Books: Novel Deep Learning Approaches and Hardware Acceleration using Intelligent Processing Units by Zihao Zhang and Stefan Zohren.
In every notebooks is proposed all the machine learning pipeline, starting from the loading of the dataset, passing from the labeling method, creation of the datasets and dataloaders, ending with the train, validation and test.
I reached the same results of the original paper.
To run the code you just have to download the FI-2010 dataset and change the data path, then the notebook will do the rest, including the training and testing.