RS学习1

最新推荐文章于 2024-05-09 09:40:31 发布

DM_LJY

最新推荐文章于 2024-05-09 09:40:31 发布

阅读量277

点赞数

分类专栏：算法整理文章标签：学习 pytorch 人工智能

本文链接：https://blog.csdn.net/DM_LJY/article/details/125288544

版权

算法整理专栏收录该内容

3 篇文章 0 订阅

订阅专栏

本文介绍了如何通过git clone和python setup.py install的方式安装torch-rechub库，并详细讲解了FM算法模型的实现。重点在于FM模型的构造和forward函数，适合深度学习推荐系统的初学者。

摘要由CSDN通过智能技术生成

torch-rechub安装

#最新版（推荐）
1. git clone https://github.com/datawhalechina/torch-rechub.git
2. cd torch-rechub
3. python setup.py install

安装过程中遇到问题：python setup.py install

执行 python3 setup.py install可以成功安装

深度学习推荐算法框架

FM算法总结

class FMModel(nn.Module):

def __init__(self, word_counts, embed_dim):
super(FMModel, self).__init__()
self.embedding = nn.Embedding(word_counts, embed_dim)
self.weight = nn.Embedding(word_counts, 1)

def forward(self, features, index):
# features: [64,39], embed:[64,39,16], weight:[64,39,1]
index = torch.tensor(index)
weights = self.weight(index)
embed = self.embedding(index)
first_embed = torch.mul(weights, features.reshape(weights.shape))
# first_embed = first_embed.squeeze(2)
# first_sum: [batch_size, 1]
first_sum = torch.sum(first_embed, dim=1, keepdim=False)
fm_embed = torch.mul(embed, features.reshape(weights.shape))
square_sum = torch.pow(torch.sum(fm_embed, dim=1, keepdim=True), 2)
sum_square = torch.sum(fm_embed * fm_embed, dim=1, keepdim=True)
# cross_term: [batch_size, 1, embed_dim]
cross_term = square_sum - sum_square
# cross_sum: [batch_size, 1]
cross_sum = 0.5 * torch.sum(cross_term, dim=2, keepdim=False)
return first_sum + cross_sum