# -*- coding: utf-8 -*-
# @Time : 2020/9/15 10:57
# @Author : Zihan Lin
# @Email : linzihan.super@foxmail.com
# @File : fnn.py
r"""
FNN
################################################
Reference:
Weinan Zhang1 et al. "Deep Learning over Multi-field Categorical Data" in ECIR 2016
"""
import torch.nn as nn
from torch.nn.init import xavier_normal_, constant_
from recbole.model.abstract_recommender import ContextRecommender
from recbole.model.layers import MLPLayers
[docs]class FNN(ContextRecommender):
"""FNN which also called DNN is a basic version of CTR model that use mlp from field features to predict score.
Note:
Based on the experiments in the paper above, This implementation incorporate
Dropout instead of L2 normalization to relieve over-fitting.
Our implementation of FNN is a basic version without pretrain support.
If you want to pretrain the feature embedding as the original paper,
we suggest you to construct a advanced FNN model and train it in two-stage
process with our FM model.
"""
def __init__(self, config, dataset):
super(FNN, self).__init__(config, dataset)
# load parameters info
self.mlp_hidden_size = config["mlp_hidden_size"]
self.dropout_prob = config["dropout_prob"]
size_list = [
self.embedding_size * self.num_feature_field
] + self.mlp_hidden_size
# define layers and loss
self.mlp_layers = MLPLayers(
size_list, self.dropout_prob, activation="tanh", bn=False
) # use tanh as activation
self.predict_layer = nn.Linear(self.mlp_hidden_size[-1], 1, bias=True)
self.sigmoid = nn.Sigmoid()
self.loss = nn.BCEWithLogitsLoss()
# parameters initialization
self.apply(self._init_weights)
def _init_weights(self, module):
if isinstance(module, nn.Embedding):
xavier_normal_(module.weight.data)
elif isinstance(module, nn.Linear):
xavier_normal_(module.weight.data)
if module.bias is not None:
constant_(module.bias.data, 0)
[docs] def forward(self, interaction):
fnn_all_embeddings = self.concat_embed_input_fields(
interaction
) # [batch_size, num_field, embed_dim]
batch_size = fnn_all_embeddings.shape[0]
output = self.predict_layer(
self.mlp_layers(fnn_all_embeddings.view(batch_size, -1))
)
return output.squeeze(-1)
[docs] def calculate_loss(self, interaction):
label = interaction[self.LABEL]
output = self.forward(interaction)
return self.loss(output, label)
[docs] def predict(self, interaction):
return self.sigmoid(self.forward(interaction))