This repository provides the official PyTorch implementation of the following paper:

XCalBid: Robust and Efficient Real-Time Bidding via Exploratory Pre-training and Calibrated Q-Learning
Authors of the paper
Affiliation withheld for anonymity

This project is built upon the hzn666/RLBid_EA codebase. Before running this repository, please ensure that you have completed the following setup steps:

git clone https://github.com/hzn666/RLBid_EA.git

Before proceeding, run the following sections of the original repository to prepare data and models:

• Data
• CTR Prediction
• Bidding Strategy (LIN)

Please follow the instructions in the original repository's documentation to execute these sections.

After completing the above setup, copy or clone this repository into the specified location within the RLBid_EA directory structure:

RLBid_EA/
├── data/                         # Data section outputs
├── lin/                          # LIN bidding strategy files
│   └── ... 
└── XCalBid/                      # Place this repository here
    ├── run.sh                    # Execution script
    └── ...                       # Other source files

We recommend using a Conda environment for dependency management:

conda create -n rlbid python=3.10.15
conda activate rlbid
conda install pytorch pytorch-cuda=11.8 -c pytorch -c nvidia
conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
pip install -r requirements.txt

To generate augmented offline data, run the following script:

bash collect-offline-data.sh

To train and test the XCalBid model, simply run:

bash run.sh

This script will execute the full training and evaluation pipeline as described in the paper.