# DL_CTR
### 注：data文件夹用来存放特征矩阵，后续的DeepFM、FM等模型需要使用GBDT_LR文件夹内feature_engineer文件夹处理的特征矩阵。
1、数据集下载：[科大讯飞初赛数据集下载](http://www.dcjingsai.com/common/cmpt/2018%E7%A7%91%E5%A4%A7%E8%AE%AF%E9%A3%9EAI%E8%90%A5%E9%94%80%E7%AE%97%E6%B3%95%E5%A4%A7%E8%B5%9B_%E7%AB%9E%E8%B5%9B%E4%BF%A1%E6%81%AF.html)
2、GBDT_LR模型：将连续特征通过GBDT离散化并结合原类别特征，使用CSR矩阵存储One-Hot以后的特征矩阵，进而用LR模型训练。
3、DeepFM模型参考[DeepFM参考链接](https://github.com/ChenglongChen/tensorflow-DeepFM)，其中FM模型的实现只需要将DeepFM中的Deep部分去掉即可实现。安利一个推荐系统的博客，[推荐系统博客](https://www.jianshu.com/p/6f1c2643d31b)。本人关于[DeepFM的博客](https://blog.csdn.net/buwei0239/article/details/86767737)。
4、Deep & Cross Network(DCN)模型参考DeepFM，实现EarlyStopping、L2正则化、Dropout，详见DCN文件夹。[DCN的博客](https://blog.csdn.net/buwei0239/article/details/86767378)。
5、Product-based Neural Networks(PNN)模型参考DeepFM，实现EarlyStopping、Dropout、inner-product和outer-product，通过use_inner参数来调用，详见PNN文件夹。[PNN的博客](https://blog.csdn.net/buwei0239/article/details/86755998)。
6、Neural Factorization Machines(NFM)模型参考DeepFM，实现EarlyStopping、Dropout，详见NFM文件夹。[NFM的博客](https://blog.csdn.net/buwei0239/article/details/86757381)
7、Attentional Factorization Machines(AFM)模型参考NFM，详见AFM文件夹。[AFM的博客](https://blog.csdn.net/buwei0239/article/details/86766008)
### 注：NFM、DeepFM训练模型速度较快；
基于科大讯飞AI营销算法比赛实现CTR深度学习方法.zip
版权申诉
168 浏览量
2024-03-28
19:57:16
上传
评论
收藏 6.11MB ZIP 举报
基于科大讯飞AI营销算法比赛实现CTR深度学习方法.zip （66个子文件） 
ignore4134 DCN 
DCN_main.ipynb 12KB metrics.py 576B DCN.py 17KB output DCN_loss0.4318.csv 2.09MB DataLoader.py 3KB __pycache__ DataLoader.cpython-36.pyc 2KB DCN.cpython-36.pyc 10KB config.cpython-36.pyc 1KB metrics.cpython-36.pyc 906B .ipynb_checkpoints DCN_main-checkpoint.ipynb 12KB config.py 2KB PNN metrics.py 576B main.ipynb 25KB PNN.py 16KB DataReader.py 3KB __pycache__ PNN.cpython-36.pyc 10KB DataReader.cpython-36.pyc 2KB config.cpython-36.pyc 1KB metrics.cpython-36.pyc 906B .ipynb_checkpoints Untitled-checkpoint.ipynb 72B main-checkpoint.ipynb 25KB config.py 2KB .gitattributes 66B data 
Readme.txt 370B DeepFM & FM metrics.py 576B main.ipynb 27KB output FM_loss0.4354.csv 2.09MB DeepFM_loss0.4311.csv 2.09MB FM_loss0.4411.csv 2.09MB DeepFM_loss0.4320.csv 2.09MB DataReader.py 3KB __pycache__ DeepFM.cpython-36.pyc 11KB DataReader.cpython-36.pyc 2KB config.cpython-36.pyc 1KB metrics.cpython-36.pyc 909B .ipynb_checkpoints main-checkpoint.ipynb 27KB DeepFM.py 17KB config.py 2KB NFM metrics.py 576B main.ipynb 26KB output nfm_loss0.4302.csv 2.09MB NFM.py 15KB DataReader.py 3KB __pycache__ DataReader.cpython-36.pyc 2KB NFM.cpython-36.pyc 9KB config.cpython-36.pyc 1KB metrics.cpython-36.pyc 906B .ipynb_checkpoints Untitled-checkpoint.ipynb 72B main-checkpoint.ipynb 26KB config.py 2KB .ipynb_checkpoints DeepFM-checkpoint.ipynb 72B README.md 2KB GBDT_LR GBDT+LR.ipynb 9KB feature_engineering.ipynb 30KB .ipynb_checkpoints feature_engineering-checkpoint.ipynb 30KB GBDT+LR-checkpoint.ipynb 9KB AFM metrics.py 576B AFM.py 15KB main.ipynb 25KB DataReader.py 3KB __pycache__ AFM.cpython-36.pyc 9KB DataReader.cpython-36.pyc 2KB config.cpython-36.pyc 1KB metrics.cpython-36.pyc 906B .ipynb_checkpoints main-checkpoint.ipynb 25KB config.py 2KB资源评论
