Improving Ad Click Prediction by Considering Non-displayed Events

Reproduce usage

Go to yh.data/split-data folder. Downloads Yahoo music R3 and extract files to that folder. (Yahoo! Music ratings for User Selected and Randomly Selected songs, version 1.0 (1.2 MB) at https://webscope.sandbox.yahoo.com/catalog.php?datatype=r )

• Install intel mkl https://software.intel.com/en-us/get-started-with-mkl-for-linux
• source source-blas.sh with intel mkl path

vim source-blas.sh
# change 'intel_mkl' path variable

• Alias python as python2.xx

alias python=python2.xx

• ./exp.sh

Concept of Each Folder

data generation

We first need to convert Yahoo Music data.

cd yh.data
vim README.md

preprocess

After we have converted data, some files should be generated before training.

cd preprocess
vim README.md

code

After having neccesary files and converted data, we can train and conduct performance evaluation by grid.sh & do-test.sh

cd code
vim README.md