Part 4. Reinforcement Learning - Q-Learning

Now we have everything set up and the basic random policy. Mario would have to be super lucky to get anywhere with this. It's time to improve the policy.

I'm going to use the reinforcement learning technique of deep Q-learning.

Deep Q-Learning

This is the …

Continue reading

Part 3. Build an api for the environment

Now that we can pull out information from the gameboy environment, why do we need to do this?

Reinforcement learning overview

I'll discuss the basic premise of reinforcement learning (RL). RL mimics how we humans learn things in the real world. We …

Continue reading

Part 2. Explore the mario environment

This is were things get exciting. I found a really well documented, well written, python based gameboy emulator - PyBoy

I fired it up and all of the games I tried worked perfectly.

Next step is to have a play with its api. There is …

Continue reading

Part 1. Build an emulator

Inspired by various videos of a computer playing game I wanted to see if I could create one for myself.
Keeping things simple the goal would be:

Goal: Create a system that learns to play super mario on the gameboy

This would need a couple …

Continue reading

Using machine learning estimate whether someone is going to commit a crime in the future.

My annotated entry is here or to play with it and run the code in google colab here

Original competition

Update

I haven't updated the site in a while but I have been busy. At …

Continue reading