SBR-Merlin

Source:
Building Session-Based Recommendation Models with Merlin [DLIT51219]

Notes:

• This repo contains the lecture from Nvidia DLI (Deep Learning Institute).
  https://courses.nvidia.com/
• Using the RAPIDS Dask-cuDF, customization required.

Original Author:
Ronay Ak, Senior Data Scientist, Nvidia
Sara Rabhi, Senior Research Scientist, Nvidia
Benedikt Schifferer, Deep Learning Engineer, Nvidia

Modified by:
Renan Monteiro Barbosa

Learning objectives:
(1) The main concepts and algorithms for SBR;
(2) How to process the data and create sequential features;
(3) How to create an SBR model with a simple MLP architecture first, then with an RNN-based architecture, and finally with a Transformer-based one using NVIDIA Merlin; and
(4) How to train/evaluate the models on GPU.

The tutorial consists of 5 notebooks:

• 01-Data-analysis-and-preparation
• 02-ETL-with-NVTabular
• 03-Next-item-prediction-with-MLP
• 04-Next-item-prediction-with-LSTM
• 05-Next-item-prediction-with-Transformers

Appendix:

Booking.com ml-dataset-mdt
https://github.com/bookingcom/ml-dataset-mdt