图编码器和解码器

图编码器和解码器原理

python实现

import torch
import torch.nn as nn
import torch.nn.init as init
import torch.nn.functional as F

class StackGCNEncoder(nn.Module):
    '''基于拼接的编码器'''
    def __init__(self, input_dim, output_dim, num_support, use_bias=False, activation=F.relu):
        '''
        对得到的分类评分使用级联的方式进行聚合
        :param input_dim {int}: 输入特征的维度
        :param output_dim {int}: 输出的特征维度，需要满足output_dim % num_support = 0
        :param num_support {int}: 评分的类别数，比如1~5分，值为5
        :param use_bias {bool, optional}: 是否使用偏置项，默认False
        :param activation {optional}: 激活函数，默认使用F.relu
        '''

        super(StackGCNEncoder, self).__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim
        self.num_support = num_support
        self.use_bias = use_bias
        self.activation = activation

        assert output_dim % num_support == 0

        self.weight = nn.Parameter(torch.Tensor(input_dim, output_dim))

        if self.use_bias:
            self.bias = nn.Parameter(torch.Tensor(output_dim,))
        self.reset_parameters()
        self.weight = self.weight.view(input_dim, output_dim//num_support, num_support)

    def reset_parameters(self):
        init.kaiming_uniform_(self.weight)
        if self.use_bias:
            init.zeros_(self.bias)

    def forward(self, user_supports, item_supports, user_inputs, item_inputs):
        '''
        :param user_supports {list of torch.sparse.FloatTensor}:
        归一化后每个评分等级对应的用户与商品邻接矩阵
        :param item_supports {list of torch.sparse.FloatTensor}:
        归一化后每个评分等级对应的商品与用户邻接矩阵
        :param user_inputs {torch.Tensor}: 用户特征的输入
        :param item_inputs {torch.Tensor}: 商品特征的输入
        :return:
            [torch.Tensor]: 用户的隐藏层特征
            [torch.Tensor]: 商品的隐藏层特征
        '''

        assert len(user_supports) == len(item_supports) == self.num_support

        user_hidden = []
        item_hidden = []

        for i in range(self.num_support):
            tmp_u = torch.matmul(self.weight[..., i], user_inputs)
            tmp_v = torch.matmul(self.weight[..., i], item_inputs)
            tmp_user_hidden = torch._sparse_mm(user_supports[i], tmp_v)
            tmp_item_hidden = torch._sparse_mm(item_supports[i], tmp_u)

            user_hidden.append(tmp_user_hidden)
            item_hidden.append(tmp_item_hidden)

        user_hidden = torch.cat(user_hidden, dim=1)
        item_hidden = torch.cat(item_hidden, dim=1)

        user_outputs = self.activation(user_hidden)
        item_outputs = self.activation(item_hidden)

        if self.use_bias:
            user_outputs += self.bias
            item_outputs += self.bias_item

        return user_outputs, item_outputs

class SumGCNEncoder(nn.Module):
    '''基于求和的编码器'''
    def __init__(self, input_dim, output_dim, num_support, use_bias=False, activation=F.relu):
        '''
        对得到的每类评分使用求和的方式进行聚合
        :param input_dim {int}: 输入特征的维度
        :param output_dim {int}: 输出的特征维度，需要满足output_dim % num_support = 0
        :param num_support {int}: 评分的类别数，比如1~5分，值为5
        :param use_bias {bool, optional}: 是否使用偏置项，默认False
        :param activation {optional}: 激活函数，默认使用F.relu
        '''
        super(SumGCNEncoder, self).__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim
        self.num_support = num_support
        self.use_bias = use_bias
        self.activation = activation

        assert output_dim % num_support == 0

        self.weight = nn.Parameter(torch.Tensor(input_dim, output_dim))

        if self.use_bias:
            self.bias = nn.Parameter(torch.Tensor(output_dim, ))
        self.reset_parameters()
        self.weight = self.weight.view(input_dim, output_dim // num_support, num_support)

    def reset_parameters(self):
        init.kaiming_uniform_(self.weight)
        if self.use_bias:
            init.zeros_(self.bias)

    def forward(self, user_supports, item_supports, user_inputs, item_inputs):
        '''
        :param user_supports {list of torch.sparse.FloatTensor}:
        归一化后每个评分等级对应的用户与商品邻接矩阵
        :param item_supports {list of torch.sparse.FloatTensor}:
        归一化后每个评分等级对应的商品与用户邻接矩阵
        :param user_inputs {torch.Tensor}: 用户特征的输入
        :param item_inputs {torch.Tensor}: 商品特征的输入
        :return:
            [torch.Tensor]: 用户的隐藏层特征
            [torch.Tensor]: 商品的隐藏层特征
        '''
        assert len(user_supports) == len(item_supports) == self.num_support

        user_hidden = 0
        item_hidden = 0

        for i in range(self.num_support):
            tmp_u = torch.matmul(self.weight[..., i], user_inputs)
            tmp_v = torch.matmul(self.weight[..., i], item_inputs)
            tmp_user_hidden = torch._sparse_mm(user_supports[i], tmp_v)
            tmp_item_hidden = torch._sparse_mm(item_supports[i], tmp_u)

            user_hidden += tmp_user_hidden
            item_hidden += tmp_item_hidden

        user_outputs = self.activation(user_hidden)
        item_outputs = self.activation(item_hidden)

        if self.use_bias:
            user_outputs += self.bias
            item_outputs += self.bias_item

        return user_outputs, item_outputs

class FullyConnected(nn.Module):
    '''非线性变换的编码器'''
    def __init__(self, input_dim, output_dim, num_support, use_bias=False, activation=F.relu, share_weights=False):
        '''
        对得到的每类评分使用非线性变换的方式进行聚合
        :param input_dim {int}: 输入特征的维度
        :param output_dim {int}: 输出的特征维度，需要满足output_dim % num_support = 0
        :param num_support {int}: 评分的类别数，比如1~5分，值为5
        :param use_bias {bool, optional}: 是否使用偏置项，默认False
        :param activation {optional}: 激活函数，默认使用F.relu
        :param share_weights {bool, optional}: 是否共享权重，默认False
        '''
        super(FullyConnected, self).__init__()

        self.input_dim = input_dim
        self.output_dim = output_dim
        self.use_bias = use_bias
        self.activation =activation
        self.share_weights = share_weights
        self.linear_user = nn.Linear(input_dim, output_dim, bias=use_bias)
        if self.share_weights:
            self.linear_item = self.linear_user
        else:
            self.linear_item = nn.Linear(input_dim, output_dim, bias=use_bias)

    def forward(self, user_inputs, item_inputs):
        '''
        前向传播
        :param user_inputs {torch.Tensor}: 输入的用户特征
        :param item_inputs {torch.Tensor}: 输入的商品特征
        :return:
            [torch.Tensor]: 输出的用户特征
            [torch.Tensor]: 输出的商品特征
        '''

        user_outputs = self.linear_user(user_inputs)
        item_outputs = self.linear_item(item_inputs)
        if self.activation:
            user_outputs = self.activation(user_outputs)
            item_outputs = self.activation(item_outputs)

        return user_outputs, item_outputs

class Decoder(nn.Module):
    def __init__(self, input_dim, num_classes):
        '''
        解码器
        :param input_dim {int}: 输入的特征维度
        :param num_classes {int}: 评分级别总数，eg.5
        '''
        super(Decoder, self).__init__()
        self.input_dim = input_dim
        self.num_classes = num_classes
        weights = []
        for i in range(self.num_classes):
            weight = nn.Parameter(torch.Tensor(input_dim, input_dim))
            weights.append(weight)

        self.reset_parameters()

    def reset_parameters(self):
        for weight in self.weights:
            init.kaiming_uniform_(weight)

    def forward(self, user_inputs, item_inputs, user_indices, item_indices):
        '''
        计算非归一化的分类输出
        :param user_inputs {torch.Tensor}: 用户的隐藏层特征
        :param item_inputs {torch.Tensor}: 商品的隐藏层特征
        :param user_indices {torch.LongTensor}: 所有交互行为中用户的id索引，与对应的item_indices构成一条边，shape=(num_edges, )
        :param item_indices {torch.LongTensor}: 所有交互行为中用户的id索引，与对应的user_indices构成一条边，shape=(num_edges, )
        :return:
            [torch.Tensor]: 未归一化的分类输出，shape=(num_edges, num_classes)
        '''

        user_inputs = user_inputs[user_indices]
        item_inputs = item_inputs[item_indices]
        outputs = []
        for weight in self.weights:
            tmp = torch.matmul(user_inputs, weight)
            out = tmp * item_inputs
            outputs.append(out)

        outputs = torch.cat(outputs, dim=1)
        return outputs