Know Your Enemies and Know Yourself in the Real-Time Bidding Function Optimisation

The code for supporting the paper under review.

Datasets

• iPinYou. We removed the feature IP and advertiser in the feature list in file make-ipinyou-data/python/mkyzx.py. Please replace the mkyzx.py file in the original make-ipinyou-data/python/ with the one provided by us, which is in the data/make-ipinyou-you-data/python/ folder. Then follow the same data-preprocess procedure on this page.

Prepare the dataset

• Prepare iPinYou dataset as described here and put make-ipinyou-data folder in the same parent folder as anonymous-submission project.
• Copy the info.txt file from the info/2259/ folder to data/make-ipinyou-data/2259/

|-- anonymous-submission
----|-- data
--------|-- make_ctr_ipinyou_dataset.py
--------|-- make-ipinyou-data
------------|-- 2259
------------...
----|-- example
----|-- marl_bidding
----|-- market_modelling
----|-- master_config.py

Installing Spinning Up

Install Spinning Up by following the instructions here

Train the CTR model

Run the make_ctr_ipinyou_dataset.py script in the data folder to generate data files with CTR values for each bid log which is stored as: train.ctr.txt

python make_ctr_ipinyou_dataset.py

Run the experiments

Single DDPG agent vs. Linear agents

./run_single_ddpg_vs_linear.sh

Multiple DDPG agents

./run_multi_ddpg.sh