NFM-pytorch

A pytorch implementation for Neural Factorization Machine (NFM) at SIGIR 2017. The original tensorflow implementation can be found at Xiangnan's repo.

Please download the dataset from here.

Performance Comparison

I run the model for 100 epochs and compare the performance shown in Table 3 of the original paper and keep all the settings identical with the original implementation (i.e., one hiddent layer, relu as the activation function, lr is 0.05 (should be), batch_size is 128 for frappe, 4096 for movielens).

Models	Frappe-128	Frappe-256	MovieLens-128	MovieLens-256
NFM-tf	0.313	0.310	0.456	0.444
NFM-pytorch	0.310	0.310	0.456	0.446

The requirements are as follows:

* python==3.6
* numpy==1.16.2
* pytorch==1.0.1
* tensorboardX==1.6 (mainly useful when you want to visulize the loss, see https://github.com/lanpa/tensorboard-pytorch)

Example to Run:

python main.py --batch_size=128 --lr=0.05 --hidden_factor=128