Training a 2048 Agent in Python

Hi Adelaide Pythonistas,

I am excited to announce September 2026 meeting!!

2048 looks simple. Slide, merge, hit 2048. Getting a computer to play it well is harder, because the board drops a random tile after every move and the obvious approaches plateau fast.

This talk builds an agent that averages over 200,000 points using four lookup tables and no neural network. It taught itself by playing 5 million games, and nobody told it a single rule about strategy. Half the talk is how that works. The other half is how I took the training run from six days down to four hour

Live demo at the end. No reinforcement learning background needed.

Speaker:

Max Huynh: is a machine learning engineer with eight years across ML, MLOps and data engineering, including five years at Agoda building production ML for one of the world's largest travel platforms. He has a Master of Computer Science from NYCU Taiwan and published multi-agent reinforcement learning work at the IJCAI 2024 Game AI Workshop. He now consults for Vocare, an Adelaide startup building speech models for pronunciation coaching, and helps organise GDG Adelaide.

Where:
University of Adelaide
Room: Bar Smith South 2040

When:
September 24th (Thursday) at 6:00 PM.

Event Sponsor: nexwave

Don't miss the opportunity to learn new things, make friends and networking.

Confirm quick as seats can go fast. 🚀

Regards

Pedro.