Skip to content
The code for 2019 Tencent College Algorithm Contest, and the online result ranks 1st in the preliminary.
Branch: master
Clone or download
Latest commit 9fa1544 Jul 18, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
models modify model and hparam Jul 11, 2019
picture update Jul 10, 2019
src remove someting Jul 10, 2019
ctrNet.py first commit Jul 10, 2019
guide.pdf first commit Jul 10, 2019
readme.md Update readme.md Jul 18, 2019
train.py modify model and hparam Jul 11, 2019

readme.md

1. 题目介绍

请直接查看guide.pdf了解赛题,该项目是初赛第一名的模型。

2.模型介绍

avatar

avatar

avatar

3. 配置环境

  • scikit-learn
  • tqdm
  • pandas
  • numpy
  • scipy
  • tensorFlow=1.12.0 (其他版本≥1.4且不等于1.5或1.6)
  • Linux Ubuntu 16.04, 128G内存(64G应该足够),一张显卡

4.数据下载

mkdir data 
cd data
#Download data from https://pan.baidu.com/s/1ASQMms_u70psRgW_KEyT2Q 
#Password: burw
unzip algo.qq.com_641013010_testa.zip imps_log.zip user.zip
cd ..

5.数据预处理

python src/preprocess.py

6.提取特征

python src/extract_feature.py

7.转换数据格式

python src/convert_format.py

1)缺失值NA用0填充

2)将Word2Vec和DeepWalk得到的embedding拼接起来,并且掩盖到5%的广告

3)将需要用key-values的稠密特征正则化到[0,1]之间

8.训练模型

mkdir submission
python train.py
You can’t perform that action at this time.