This is a demonstration for the experiments in the paper "Budget-Constrained Real-Time Bidding Optimization: Multiple Predictors Make It Better".
- R version >= 3.4.4 with the following packages installed:
magrittr
dplyr
data.table
Matrix
FeatureHashing
stringr
- Python 2.7
- Clone this repository into your home directory (
~/TKDD2019/) - In
~/TKDD2019/db.data/CV1/,tar zxvf 2259-1.tgz; tar zxvf 2259-2.tgz - In
~/TKDD2019/db.data/, run./CVsimple.sh
AD: 2259 CV1
Benchmark ADID = 2259, BDGT_DIV = 64
$Total.Cost
[1] 19451381
$Total.Clk
[1] 69
$Total.Clk.Price
[1] 8593
$Campaign.Budget
[1] 303927.8
$Predicted.Clk
[1] 0
PERFECT: Clk 69 Cost 8593 BLeft 295334.83
ORTB C = 50.383358
ORTB with lambda 6.500160E-06: Clk 12 Cost 303923 BLeft 4.83
ORTB with lambda 6.500162E-06: Clk 12 Cost 303923 BLeft 4.83
PRUD2 lift = 144
PRUD2 lift 144, bf 5.387279E-06: nImp 201440, Clk 18 Cost 303790 BLeft 137.83
PRUD2 lift 144, bf 5.387280E-06: nImp 201440, Clk 18 Cost 303790 BLeft 137.83
This project reuses a part (namely, CTR predictor) from the work of Weinan Zhang (https://github.com/wnzhang/optimal-rtb) under the Apache 2.0 license.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent