Reproduce the classic and open source ctr model
复现了一些经典开源ctr模型，并持续完善
The code heavily references the Deepctr project for learning purposes only
代码大量参考了DeepCTR项目，仅做学习所用
DeepCTR URL：https://github.com/shenweichen/DeepCTR-Torch
The reproduction performance did not meet the results claimed in the paper
复现性能没有达到论文中宣称的结果
Possible reasons for code bugs or inconsistent data
可能是代码存在Bug或者数据不一致的原因
已复现模型（Reproduced model）：
| 模型名（model name） | 论文地址（paper address） |
|---|---|
| Wide&Deep | https://arxiv.org/pdf/1606.07792v1.pdf |
| DeepFM | https://arxiv.org/pdf/1703.04247v1.pdf |
| xDeepFM | https://arxiv.org/pdf/1803.05170v3.pdf |
| DCN | https://arxiv.org/pdf/1708.05123.pdf |
| AutoInt | https://arxiv.org/pdf/1810.11921v2.pdf |
| AFN | https://arxiv.org/pdf/1909.03276v2.pdf |
| GateNet | https://arxiv.org/pdf/2007.03519v1.pdf |
| FiBiNet | https://arxiv.org/pdf/1905.09433v1.pdf |
| FATDeepFFM | https://arxiv.org/pdf/1905.06336v1.pdf |
| FiBiNetPlus | https://arxiv.org/pdf/2209.05016v1.pdf |
| ContextNet | https://arxiv.org/pdf/2107.12025v1.pdf |
| DCNv2 | https://arxiv.org/pdf/2008.13535v2.pdf |
| MaskNet | https://arxiv.org/pdf/2102.07619v2.pdf |
| FinalMLP | https://arxiv.org/pdf/2304.00902v3.pdf |
运行流程 running process
第一步 step1
数据准备 prepare training data
$ cd data/criteo
$ python download_criteo_x1.py
$ python trans.py (please modify the path accordingly)第二步 step2
提前在config中配置好需要的模型和参数 Configure the required models and parameters in advance in config
第三步 step3
在main.py中配置模型设置和参数设置 configure model and parameter settings in main.py
以FinalMLP为例 Taking FinalMLP as an example
$ python mian.py --config_dir='./config/FinalMLP_criteo_x1' --model_setid='base' data_setid='base' gpu_index=0 expid='v1'