DeepFM

Deep Factorization Machine Model for CRT prediction.

References

• DeepFM: A Factorization-Machine based Neural Network for CTR Prediction (Paper)
• Factorization Machines
• Sistemas de Recomendación (Parte 1): Filtros Colaborativos | Clase 22 | Aprendizaje Profundo 2021
• Factorization Machine models in PyTorch

Notebooks

• Use example
• rs-check-how-works

Requisites

• anaconda / miniconda
• pytorch-common
  • Github repo
  • Pypi repo

Getting started

Step 1: Clone repo.

$ git clone https://github.com/adrianmarino/deep-fm.git
$ cd deep-fm

Step 2: Create environment.

$ cd dfm
$ conda env create -f environment.yml

Step 3: Enable project environment.

$ conda activate deepfm

Step 3: Run regression tests.

$ pytest

Training

$ python bin/train

$ python bin/train --help

Usage: train [OPTIONS]

Options:
  --device TEXT                   Device used to functions and optimize model.
                                  Values: gpu(default) or cpu.
  --cuda-process-memory-fraction FLOAT
                                  Setup max memory used per CUDA process.
                                  Percentage expressed between 0 and
                                  1(default: 0.5).
  --dataset TEXT                  Select movie lens dataset type. Values:
                                  1m(default), 20m.
  --cv-n-folds INTEGER            cross validation n folds(default: 10).
  --train-percent FLOAT           Observations percent to used on training
                                  process(default: 0.7).
  --help                          Show this message and exit.