Article

← Retour au blog

🇫🇷 Lire en français

August 2005: A machine learning 3D app to crack the lotto

2005-08-25 #903 · rév. 1 · EN

During the summer of 2005, I spent a month working on a way to improve my chances of winning the French lottery, and in August, I came up with SdlLoto.

SdlLoto: A real-time 3D visualization with WarStation™ / Java

I used the WarStation port to Java that I completed in July to develop SdlLoto.

I thought I could find a 3D visualization that would help me develop the ideas I had.

Failing to find visual patterns

I initially thought I could find visual patterns in series of winning numbers, but I got nowhere.

Machine learning: Neural network and DNA-programming

Several years earlier, I had explored genome programming on my HP4/CV in order to recognize letter glyphs.

For the lotto, I wanted to simulate multiple ways to draw 6 numbers and test them against all the previous actual winning draws I had in my database.

I created millions of players, each picking a drawing method, and generated 1024 generations. Because the computing time was long, I made a server version of the software and ran it on a server overnight. Once the best guess was ready, I could run the 3D client version on my computer, get the predicted numbers, and play them for real.

In order to help trace the whole process, I visualized the multiple genomes as a rolling bitmap image which I displayed at the bottom of the screen:

My conclusion was that I really thrived working on such a subject and learned tons of things. But I never won the lottery, though. No problem.