Reproduce selected results from Transformers4Rec paper with Merlin Models API

[gabrielspmoreira]

Description

For Transformers4Rec, we have created a training/eval script for reproducing the paper experiments, that takes a set of hparams as command line arguments and a preprocessed dataset.

This task is about creating the training/eval script based on the original Transformers4Rec script using the Merlin Models API, which will:

• Ensure that our TF implementation is correct and that it matches the results we had using Transformers4Rec (PyTorch)
• Work as an advanced example on how to set the available hparams for session-based recommendation

Selected best results to reproduce with REES46 dataset (without features):

• GPT-2 (CLM)
• XLNet (CLM)
• XLNet (MLM)
• XLNet - ALL Features (MLM)

We should compare accuracy and runtimes for best trials, reported in this spreadsheet, in the paper and in the paper online appendix