最近在看一篇Google的有關(guān)CTR預估的paper：Deep & Cross Network for Ad Click Predictions
整理記錄一下參考的資料和自己的部分體會。
不錯的參考資料
Deep and Cross Network原理及實現(xiàn)（有原理有代碼）
Wide & Deep Learning for Recommender Systems Google的另一篇推薦系統(tǒng)的類似paper
Keras中文文檔
[GitHub | DeepCTR] (https://github.com/shenweichen/DeepCTR)
遇到的問題
Deep and Cross Network原理及實現(xiàn)
- keras.layers.merge報錯，需要修改為concatenate
from keras.layers import merge
input = merge(inp_embed, mode = 'concat')
output = merge([deep, cross], mode = 'concat')
參考文檔：https://keras.io/layers/core/
from keras.layers import concatenate
input = concatenate(inp_embed, axis=-1)
output = concatenate([deep, cross], axis=-1)
- 畫模型圖需要安裝額外的包
plot_model(model, to_file = 'model.png', show_shapes = True)
參考：https://stackoverflow.com/questions/40632486/dot-exe-not-found-in-path-pydot-on-python-windows-7
pip install django-extensions
pip install pyparsing
pip install graphviz
pip install pydot
conda install graphviz
