DEEP & WIDE MODEL

轉自：https://www.jianshu.com/p/28a1849f6707

前言：應該隻要一天時間，先看Google英文論文，配合中文翻譯：http://d0evi1.com/widedeep-recsys，再看些官方的代碼，需要先了解下tensorflow的feature API：

tf.contrib.layers 中提供了一系列的函數定義不同類型的 feature columns：

tf.contrib.layers.sparse_column_with_XXX 建構低維離散特征

sparse_feature_a = sparse_column_with_hash_bucket(…)

sparse_feature_b = sparse_column_with_hash_bucket(…)

tf.contrib.layers.crossed_column 建構離散特征的組合

sparse_feature_a_x_sparse_feature_b = crossed_column([sparse_feature_a, sparse_feature_b], …)

tf.contrib.layers.real_valued_column 建構連續型實數特征

real_feature_a = real_valued_column(…)

tf.contrib.layers.embedding_column 建構 embedding 特征

sparse_feature_a_emb = embedding_column(sparse_id_column=sparse_feature_a, )

feature_column API：

https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/feature_column/feature_column_lib.py

1）Google 官方資料：

wide and deep論文：

https://arxiv.org/pdf/1606.07792.pdf

wide and depp 主函數：https://github.com/tensorflow/tensorflow/blob/b474e55c23e5cc42b01a1ddea34751f01110deb6/tensorflow/python/estimator/canned/dnn_linear_combined.py

tf.estimator.DNNLinearCombinedClassifier主函數參數：

https://www.w3cschool.cn/tensorflow_python/tensorflow_python-4gn82dty.html

wide_n_deep_tutorial.py代碼例子：https://github.com/tensorflow/tensorflow/blob/88abddbc30a34855d8cdf44c81eac388fbec955b/tensorflow/examples/learn/wide_n_deep_tutorial.py

wide_and_deep.md：

https://github.com/tensorflow/tensorflow/blob/752dcb61ef7a8fd6555909dc37c1f2a2e5792227/tensorflow/docs_src/tutorials/wide_and_deep.md

TensorFlow Linear Model Tutorial：

https://github.com/tensorflow/tensorflow/blob/752dcb61ef7a8fd6555909dc37c1f2a2e5792227/tensorflow/docs_src/tutorials/wide.md

2）非官網參考資料：

TensorFlow下的大規模線性模型：

http://www.jianshu.com/p/afe41ae17a16

tensorflow邏輯回歸：http://blog.csdn.net/willduan1/article/details/53105603

tensorlfow線性模型

http://shuaizhang.tech/2016/12/17/Tensorflow-Tutorial-5-Linear-Model-%E7%BA%BF%E6%80%A7%E6%A8%A1%E5%9E%8B/

wide and depp 論文中文翻譯：

基于Wide & Deep Learning的推薦系統http://d0evi1.com/widedeep-recsys/

Summary Of recommender system surveys in recent years：

http://shuaizhang.tech/2017/07/28/Summary-of-Recommender-System-Surveys-in-recent-years/

各大公司廣泛使用的線上學習算法FTRL詳解：

http://www.cnblogs.com/EE-NovRain/p/3810737.html

Tensorflow - wide and deep.ipynb：

https://github.com/bartgras/XGBoost-Tensorflow-Wide-and-deep-comparison/blob/master/Tensorflow%20-%20wide%20and%20deep.ipynb

TensorFlow Wide And Deep 模型詳解與應用：

http://geek.csdn.net/news/detail/235465

TensorFlow Wide And Deep 模型詳解與應用（二）：

http://geek.csdn.net/news/detail/235471

tensorflow線性模型以及Wide deep learning：

http://blog.csdn.net/zhangshuaizaxia/article/details/53707301

TensorFlow 廣度和深度學習教程：

http://www.jianshu.com/p/7dc588d98a94

深度學習在美團點評推薦平台排序中的應用&& wide&&deep推薦系統模型--學習筆記：

http://www.cnblogs.com/wuxiangli/p/7258474.html

Google_Wide&Deep Learning

http://www.datakit.cn/blog/2016/08/21/wdnn.html