提交 3cb4e477 编写于 作者: Y yaoxuefeng

Merge branch 'xuefeng' into 'develop'

modify rank readme

See merge request !11
# Rank模型库
# 排序模型库
## 简介
我们提供了常见的ctr任务中使用的模型,包括 [dnn](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/dnn)[dcn](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/dcn)[deepfm](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/deepfm)[xdeepfm](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/xdeepfm)[din](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/din)[wide&deep](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/wide_deep)
我们提供了常见的排序任务中使用的模型算法,包括 [多层神经网络](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/dnn)[Deep Cross Network](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/dcn)[DeepFM](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/deepfm)[xDeepFM](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/xdeepfm)[Deep Interest Network](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/din)[Wide&Deep](http://gitlab.baidu.com/tangwei12/paddlerec/tree/develop/models/rank/wide_deep)
模型算法库在持续添加中,欢迎关注。
......@@ -23,11 +23,11 @@
| 模型 | 简介 | 论文 |
| :------------------: | :--------------------: | :---------: |
| DNN | 多层神经网络 | -- |
| wide&deep | Deep + wide(LR) | [论文链接](https://dl.acm.org/doi/abs/10.1145/2988450.2988454)(2016) |
| DeepFM | Deep + FM 并行 | [论文链接](https://arxiv.org/abs/1703.04247)(2017) |
| xDeepFM | DeepFM升级版 | [论文链接](https://dl.acm.org/doi/abs/10.1145/3219819.3220023)(2018) |
| DCN | wide升级为Cross Layer Network | [论文链接](https://dl.acm.org/doi/abs/10.1145/3124749.3124754)(2017) |
| DIN | Embeddding层引入attention机制 | [论文链接](https://dl.acm.org/doi/abs/10.1145/3219819.3219823)(2018) |
| wide&deep | Deep + wide(LR) | [Wide & Deep Learning for Recommender Systems](https://dl.acm.org/doi/abs/10.1145/2988450.2988454)(2016) |
| DeepFM | Deep + FM 并行 | [DeepFM: A Factorization-Machine based Neural Network for CTR Prediction](https://arxiv.org/abs/1703.04247)(2017) |
| xDeepFM | DeepFM升级版 | [xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems](https://dl.acm.org/doi/abs/10.1145/3219819.3220023)(2018) |
| DCN | wide升级为Cross Layer Network | [Deep & Cross Network for Ad Click Predictions](https://dl.acm.org/doi/abs/10.1145/3124749.3124754)(2017) |
| DIN | Embeddding层引入attention机制 | [Deep Interest Network for Click-Through Rate Prediction](https://dl.acm.org/doi/abs/10.1145/3219819.3219823)(2018) |
## 使用教程
### 数据处理
......@@ -37,22 +37,22 @@
## 效果对比
### 模型效果列表
| 数据集 | 模型 | 单机测试集指标 | 详情 |
| :------------------: | :--------------------: | :---------: |:---------: |
| Criteo | DNN | auc:0.79395 | [更多](https://github.com/PaddlePaddle/models/tree/develop/PaddleRec/ctr/dnn#benchmark) |
| Criteo | DeepFM | logloss: 0.44797, <br>auc:0.8046 | [更多](https://github.com/PaddlePaddle/models/tree/develop/PaddleRec/ctr/deepfm#result) |
| Criteo | DCN | logloss: 0.44703564, <br>auc: 0.80654419 | [更多](https://github.com/PaddlePaddle/models/tree/develop/PaddleRec/ctr/dcn#%E7%BB%93%E6%9E%9C) |
| Demo数据集 | xDeepFM | acc: 0.48657, <br>auc:0.7308 | [更多](https://github.com/PaddlePaddle/models/tree/develop/PaddleRec/ctr/xdeepfm#%E5%8D%95%E6%9C%BA%E7%BB%93%E6%9E%9C) |
| Census-income Data | Wide&Deep | mean_acc:0.76195, <br>mean_auc:0.90577 | [更多](https://github.com/PaddlePaddle/models/tree/develop/PaddleRec/ctr/wide_deep#%E6%A8%A1%E5%9E%8B%E6%95%88%E6%9E%9C) |
| Amazon Product | DIN | logloss: 0.47005194, <br>auc: 0.863794952818 | [更多](https://github.com/PaddlePaddle/models/tree/develop/PaddleRec/ctr/din#%E9%A2%84%E6%B5%8B%E7%BB%93%E6%9E%9C%E7%A4%BA%E4%BE%8B) |
| 数据集 | 模型 | loss | 测试auc | acc | mae |
| :------------------: | :--------------------: | :---------: |:---------: | :---------: |:---------: |
| Criteo | DNN | -- | 0.79395 | -- | -- |
| Criteo | DeepFM | 0.44797 | 0.8046 | -- | -- |
| Criteo | DCN | 0.44703564 | 0.80654419 | -- | -- |
| Criteo | xDeepFM | -- | 0.7308 | 0.48657 | -- |
| Census-income Data | Wide&Deep | 0.76195(mean) | 0.90577(mean) | -- | -- |
| Amazon Product | DIN | 0.47005194 | 0.863794952818 | -- | -- |
## 分布式
### 模型性能列表
| 数据集 | 模型 | 单机 | 多机(同步) | 多机(异步) |
| :------------------: | :--------------------: | :---------: |:---------: |:---------: |
| Criteo | DNN | -- | -- | -- |
| Criteo | DeepFM | -- | -- | -- |
| Criteo | DCN | -- | -- | -- |
| Demo数据集 | xDeepFM | -- | -- | -- |
| Census-income Data | Wide&Deep | -- | -- | -- |
| Amazon Product | DIN | -- | -- | -- |
| 数据集 | 模型 | 单机 | 多机(同步) | 多机(异步) | GPU |
| :------------------: | :--------------------: | :---------: |:---------: |:---------: |:---------: |
| Criteo | DNN | -- | -- | -- | -- |
| Criteo | DeepFM | -- | -- | -- | -- |
| Criteo | DCN | -- | -- | -- | -- |
| Criteo | xDeepFM | -- | -- | -- | -- |
| Census-income Data | Wide&Deep | -- | -- | -- | -- |
| Amazon Product | DIN | -- | -- | -- | -- |
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册