Jupylet is a Python library that lets you create and run games interactively in a Jupyter notebook. It is intended for two types of audiences:

• Kids and their parents interested in learning to program,
• Researchers and students of deep reinforcement learning.

Here are two examples of 2D and 3D games for the impatient. Their code is included in this repository:

A Jupyter notebook is in essence a laboratory for programming. It is the ideal environment for playing around with code, experimenting, and exploring ideas. It is used by professional machine learning scientists who come every day to play at work, so why not by kids?

Jupylet is built on top of pyglet, a powerful game programming library. It is wonderfully easy to use for creating simple 2D games and Jupylet extends it to 3D gaming.

By marrying Jupyter and pyglet, Jupylet lets you have the best of both worlds. Create games interactively and experimentally, change a variable or a function and see how the game is affected immediately while it is running.

Jupylet makes it is super easy to create and modify environments in which to experiment with deep reinforcement learning algorithms and it includes the code required to programmatically control multiple simultaneous games and render thousands of frames per second.

Consider for example the pong game included in this repository. With a few lines of code you can modify the colors of the game or turn the background into a swirling giant flower to experiment with transfer learning, or turn the game into 4-way pong with agents on all four sides of the game court to experiment with cooperation between multiple agents. And since you can modify the game interactively in Jupyter this process is not only easy but fun.

Head to examples/22-pong-RL.ipynb to see how to programmatically control and render a 2-player version of pong.

jupylet should run on Python 3.4+ on Windows, Mac, and Linux.

If you are new to Python, I strongly recommend that you install and use the Miniconda Python distribution. Download and run the 64-bit installer and stick to the default install options.

Once Miniconda is installed start a Miniconda Prompt. To do this on Windows click the ⊞ Winkey then type Miniconda and press the Enter key. This should open a small dark window that programmers call console or shell in which you can enter commands and run programs.

To run jupylet first install it by typing the following command in the console:

pip install jupylet

Next, you need to download the jupylet repository since it contains the example notebooks. If you have Git installed you can use it to clone the jupylet repository with:

git clone https://github.com/nir/jupylet.git

Alternatively, if you don't have Git installed, you can download and unzip the jupylet archive by typing:

python -m wget https://github.com/nir/jupylet/archive/master.zip
python -m zipfile -e jupylet-master.zip .
move jupylet-master jupylet

Next, enter the jupylet/examples/ directory with the change directory command:

cd ./jupylet/examples/

And start a jupyter notebook with:

jupyter notebook 11-spaceship.ipynb

Run the notebook by following the instructions in the notebook and a game canvas should appear with the spaceship example:

You can run essentially the same code from the console with:

python spaceship.py

The only difference in the code is that the application object is instantiated with:

app = App(mode='window')

In terms of its interface jupylet introduces only minor additions and modifications to the underlying pyglet library. Therefore the bulk of learning to use it is covered by the pyglet documentation. I have started writing a guide and will add documentation in the comming days; in the mean time head to the 11-spaceship.ipynb example to get started.

For questions and feedback send an email to Nir Aides.