Combinatorial Causal Bandits

This repository contains code of numerical experiments for paper:
Combinatorial Causal Bandits
AAAI 2023
Shi Feng, Wei Chen
[ArXiv Version]

Usage

$G_1,G_2,\cdots,G_4$ are parallel binary linear models. $G_1$ and $G_2$ are shown as below:

$G_3$ is $G_2$ without $X_8$ and $X_9$. $G_4$ is $G_2$ removing $X_6,X_7,X_8$ and $X_9$. $G_5$ is a two-layer BLM shown as below:

If you want to compare regrets of BLM-OFU, BLM-LR, UCB and $\epsilon$-greedy algorithms on graph $G_*$, you need to run

python main.py --G*

You can find our running samples in ./results/ directory.

File Hierarchy

.
├── utils/                    # implementations of BLM and online algorithms
│   ├── blm_lr.py             # implementing BLM-LR algorithm
│   ├── blm_ofu.py            # implementing BLM-OFU (BGLM-OFU) algorithm
│   ├── epsilon_greedy.py     # implementing epsilon-greedy algorithm
│   ├── ucb.py                # implementing UCB algorithm
│   ├── parallel_graph.py     # implementing parallel graph (G1, G2, G3, G4)
│   └── two_layer_graph.py    # implementing two-layer graph (G5)
├── results/                  # our running samples
├── main.py                   # main file
├── .gitignore                # exclude some annoying files from git
├── LICENSE                   # MIT license
└── README.md                 # what you are reading now

Contact

If you have any questions, feel free to contact us through email (shifeng-thu@outlook.com) or Github issues. Enjoy!