Reproduce the Experienments in our paper: Deep Censored Learning of the Winning Price in the Real Time Bidding

Please install docker and use the following commands to re-run the experiments

docker run -it wush978/deepcensor:latest /bin/bash
# under docker
cd deepcensor
git pull origin master
source bin/activate
cd exp
python train.py --config linear-normal-no-ipinyou.exp.data-201310_1e-4/01.json

Each .json file under the folder exp is corresponding to one experiment in the paper. The name of the subdirectories is generated by the <link-structure>-<loss>-<censoring>-ipinyou.exp.data-201310_1e-4. Due to data size issue, only data of the iPinYou 3rd Season are released.

Reconstruct the Training Data

To reconstruct the training dataset, please visit the iPinYou Real-Time Bidding Dataset for Computational Advertising Research to download the ipinyou.contest.dataset.zip and place the file in the root directory of this project.

The reconstruction requires the following tools, their packages / modules, and their system dependencies:

• R 3.4.2
• python 3.6.3

We use linuxbrew to build these tools and the installed packages are under brew.list. Or the reader could download the docker image: wush978/deepcensor:build for a clone of our environment.

To initialize R environment, please run the following script under shell after installing R 3.4.2:

Rscript -e "install.packages('remotes')"
Rscript -e "remotes::install_github('wush978/pvm')"
Rscript -e "pvm::import.packages()"

Then, please run the following command:

make