deepctr_torch/models/difm.py

# -*- coding:utf-8 -*-
"""
Author:
    zanshuxun, zanshuxun@aliyun.com
Reference:
    [1] Lu W, Yu Y, Chang Y, et al. A Dual Input-aware Factorization Machine for CTR Prediction[C]//IJCAI. 2020: 3139-3145.(https://www.ijcai.org/Proceedings/2020/0434.pdf)
"""
import torch
import torch.nn as nn

from .basemodel import BaseModel
from ..inputs import combined_dnn_input, SparseFeat, VarLenSparseFeat
from ..layers import FM, DNN, InteractingLayer, concat_fun


class DIFM(BaseModel):
    """Instantiates the DIFM Network architecture.

    :param linear_feature_columns: An iterable containing all the features used by linear part of the model.
    :param dnn_feature_columns: An iterable containing all the features used by deep part of the model.
    :param att_head_num: int. The head number in multi-head  self-attention network.
    :param att_res: bool. Whether or not use standard residual connections before output.
    :param dnn_hidden_units: list,list of positive integer or empty list, the layer number and units in each layer of DNN
    :param l2_reg_linear: float. L2 regularizer strength applied to linear part
    :param l2_reg_embedding: float. L2 regularizer strength applied to embedding vector
    :param l2_reg_dnn: float. L2 regularizer strength applied to DNN
    :param init_std: float,to use as the initialize std of embedding vector
    :param seed: integer ,to use as random seed.
    :param dnn_dropout: float in [0,1), the probability we will drop out a given DNN coordinate.
    :param dnn_activation: Activation function to use in DNN
    :param dnn_use_bn: bool. Whether use BatchNormalization before activation or not in DNN
    :param task: str, ``"binary"`` for  binary logloss or  ``"regression"`` for regression loss
    :param device: str, ``"cpu"`` or ``"cuda:0"``
    :param gpus: list of int or torch.device for multiple gpus. If None, run on ``device`` . ``gpus[0]`` should be the same gpu with ``device`` .
    :return: A PyTorch model instance.

    """

    def __init__(self,
                 linear_feature_columns, dnn_feature_columns, att_head_num=4,
                 att_res=True, dnn_hidden_units=(256, 128),
                 l2_reg_linear=0.00001, l2_reg_embedding=0.00001, l2_reg_dnn=0, init_std=0.0001, seed=1024,
                 dnn_dropout=0,
                 dnn_activation='relu', dnn_use_bn=False, task='binary', device='cpu', gpus=None):
        super(DIFM, self).__init__(linear_feature_columns, dnn_feature_columns, l2_reg_linear=l2_reg_linear,
                                   l2_reg_embedding=l2_reg_embedding, init_std=init_std, seed=seed, task=task,
                                   device=device, gpus=gpus)

        if not len(dnn_hidden_units) > 0:
            raise ValueError("dnn_hidden_units is null!")

        self.fm = FM()

        # InteractingLayer (used in AutoInt) = multi-head self-attention + Residual Network
        self.vector_wise_net = InteractingLayer(self.embedding_size, att_head_num,
                                                att_res, scaling=True, device=device)

        self.bit_wise_net = DNN(self.compute_input_dim(dnn_feature_columns, include_dense=False),
                                dnn_hidden_units, activation=dnn_activation, l2_reg=l2_reg_dnn,
                                dropout_rate=dnn_dropout,
                                use_bn=dnn_use_bn, init_std=init_std, device=device)
        self.sparse_feat_num = len(list(filter(lambda x: isinstance(x, SparseFeat) or isinstance(x, VarLenSparseFeat),
                                               dnn_feature_columns)))

        self.transform_matrix_P_vec = nn.Linear(
            self.sparse_feat_num * self.embedding_size, self.sparse_feat_num, bias=False).to(device)
        self.transform_matrix_P_bit = nn.Linear(
            dnn_hidden_units[-1], self.sparse_feat_num, bias=False).to(device)

        self.add_regularization_weight(
            filter(lambda x: 'weight' in x[0] and 'bn' not in x[0], self.vector_wise_net.named_parameters()),
            l2=l2_reg_dnn)
        self.add_regularization_weight(
            filter(lambda x: 'weight' in x[0] and 'bn' not in x[0], self.bit_wise_net.named_parameters()),
            l2=l2_reg_dnn)
        self.add_regularization_weight(self.transform_matrix_P_vec.weight, l2=l2_reg_dnn)
        self.add_regularization_weight(self.transform_matrix_P_bit.weight, l2=l2_reg_dnn)

        self.to(device)

    def forward(self, X):
        sparse_embedding_list, _ = self.input_from_feature_columns(X, self.dnn_feature_columns,
                                                                   self.embedding_dict)
        if not len(sparse_embedding_list) > 0:
            raise ValueError("there are no sparse features")

        att_input = concat_fun(sparse_embedding_list, axis=1)
        att_out = self.vector_wise_net(att_input)
        att_out = att_out.reshape(att_out.shape[0], -1)
        m_vec = self.transform_matrix_P_vec(att_out)

        dnn_input = combined_dnn_input(sparse_embedding_list, [])
        dnn_output = self.bit_wise_net(dnn_input)
        m_bit = self.transform_matrix_P_bit(dnn_output)

        m_x = m_vec + m_bit  # m_x is the complete input-aware factor

        logit = self.linear_model(X, sparse_feat_refine_weight=m_x)

        fm_input = torch.cat(sparse_embedding_list, dim=1)
        refined_fm_input = fm_input * m_x.unsqueeze(-1)  # \textbf{v}_{x,i}=m_{x,i} * \textbf{v}_i
        logit += self.fm(refined_fm_input)

        y_pred = self.out(logit)

        return y_pred