The dino breeding farm

Welcome to our dino breeding farm. Unlike others, we do not grow dino for movie marking.
Here, we desire to breed dino in order to win the very famous cactus run ! To do so, each year,
we select our best dinos to become the breeder of a new generation of dinos. Thus, as generation
increases, our dinos become better and better.

On this page, you can challenge our best dino for each of the past generations. As, you can see
first generations was not very good, but they become better after some time. Click on stop and
choose a generation below to challenge the best of this generation. Your dino is the one below,
just jump with the space bar.

Comments

• The game is a C++ copy of the famous chromium game. It is calibrated by hand, and so can
  be harder than the original one.
• The dinos are governed by a neural network trained with genetic algorithms.
• This project has been done for the blog Briques2maths in order to show some ludic applications
  of neural networks.
• The whole source code is available on my git. It can be freely used as a base to design projects
  for students. For such purposes, let us highlight that C++ can be interfaced with Python using Boost.
  So, one could use standard machine learning libraries. Unfortunately, for now, the code is not
  commented nor documented.
• The source code contains:
  • the game,
  • a very basic neural network toolbox,
  • and an ersatz of game engine allowing to do this with other (but still very simple) games
• The whole project has been ported from C++ to javascript thanks to the amazing emscripten.
• Of course, an expert system would have easily done the job. Our goals was pedagological.