This implementation is based on the framework for CTR predictions found @ https://github.com/shenweichen/DeepCTR-Torch.
(Recommended) Requirements are: Python (3.9.12), tensorflow (2.8.0), sklearn, xgboost (1.5.2), torch (1.11.0), tqdm and pandas. (Note it might well work with other versions but it was not tested)
Brief: FM trained together with a MLPMixer for CTR prediction.
Figure 1: Full FMMixer model
Figure 2: Mixer component
Figure 3: Single mixer layer
Sample data (100K entries) is already present, to do a full run with preprocessing and training run Avazu version and Criteo version.
Note: you might need to add the project folder to PYTHONPATH, or sys.path.append() depending on your user case.