https://github.com/thchilly/blackjack-card-counting
Optimal Black Jack policy that also attempts to use card counting to beat the dealer using Reinforcement Learning techniques
https://github.com/thchilly/blackjack-card-counting
Last synced: 9 months ago
JSON representation
Optimal Black Jack policy that also attempts to use card counting to beat the dealer using Reinforcement Learning techniques
- Host: GitHub
- URL: https://github.com/thchilly/blackjack-card-counting
- Owner: thchilly
- Created: 2025-05-27T11:06:17.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-05-29T04:51:50.000Z (about 1 year ago)
- Last Synced: 2025-06-25T13:48:05.115Z (about 1 year ago)
- Language: Jupyter Notebook
- Size: 16.6 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Blackjack Card Counting - RL Project
A comprehensive Python framework for Blackjack, combining environment simulation, model-based benchmarks, model-free learning, card-counting, and a user-friendly GUI:
- **Custom Environment**
A custom `BlackjackEnv` implementation: supports multiple decks, reshuffling, natural-pay rules (3:2), ace handling, and a running Hi–Lo count.
- **Model-Based Ground Truth**
Exact MDP solutions via Value Iteration and Policy Iteration under infinite-deck assumptions, yielding the canonical “basic strategy” hit/stand policy.
- **Model-Free Baseline**
Tabular Q-Learning recovers the optimal hit/stand policy from scratch, using linearly decaying learning and exploration rates.
- **Card-Counting Extension**
Augments the Q-Learning state with a simple Hi–Lo running count (Low/Neutral/High bins) to adapt play decisions based on shoe composition.
- **Graphical Interface**
A Tkinter-based GUI for human play: choose decks, view dealer upcard and remaining cards, track session stats, and play via Hit/Stay buttons.
- **Experiments & Visualization**
Jupyter notebook with training scripts, learning curves, win-rate comparisons, and policy heatmaps for both baseline and counting agents.
- **Full Report (PDF)**
Detailed write-up of methodology, experiments, and results in `report/Blackjack_Card_Counting_Phase1.pdf`.
## Project Structure
```text
blackjack/
├── agents/
│ ├── q_learning_agent.py
│ └── value_iteration.py
├── envs/
│ ├── blackjack_env.py
│ └── blackjack_env1.py
├── gui/
│ └── simple_gui.py
├── materials/
│ └── playing-cards-master/
├── notebooks/
│ └── experiments.ipynb
├── figures/
├── test_env.py
├── requirements.txt
└── README.md
```
## Installation
1. **Clone** this repo
```bash
git clone https://github.com/yourusername/blackjack.git
cd blackjack
```
2. **Create and Activate** a virtual-env (optional but recommended)
```bash
python3 -m venv .venv
source .venv/bin/activate # Mac/Linux
.venv\Scripts\activate # Windows
```
4. **Install** dependencies
```bash
pip install -r requirements.txt
```
## Smoke Test
Run a quick environment sanity check:
```bash
python test_env.py
```
## Launch the GUI
```bash
python -m gui.simple gui
```
* Enter number of decks (1–8).
* Use the 👊 Hit, ✋ Stay, and 🔄 New Game buttons.
* Watch session stats & remaining cards update in real time.
## Jupyter Experiments and Analysis
All RL training & evaluation code is in:
```bash
notebooks/experiments.ipynb
```
Launch with:
```bash
jupyter lab notebooks/experiments.ipynb
```