Fatball learns to survive by regulating his temperature and energy. Fatball is trained on human and self play data - it uses the data points from human attempts at regulation of basic variables of survival (temperature and energy). Based on the successfull data points (surviving and obtaining 5 pieces of food) Fatball learns to self-regulate.
Fatball is made for two different motives
- To familiarise ourselves with basic machine learning methods beginning with most well known ones : decision trees, to eventually move on to deep and unsupervised learning
- To lay down the foundation for building simulation based animats
- Clone the repo
- Install
pygame,sklearn,numpy - Run the program
- The ball always start in the center at the beginning of each turn.
- 5 Foods (green dots) will be scattered randomly across the screen.
- The ball can do two operations - wait or go to the nearest food.
- There are two gauges at the top of the screen which shows - temperature and hunger, respectively
- If the ball stays still, the temperature and food will go down.
- If the ball moves, the temperature will rise and the food will go down at a higher rate than when stationary.
- If the temperature drops below 0, or goes over the limit, its game over.
- If the food drops below 0, the its game over.
- There is no limit to food gathering.
The game has two modes:
- Manual play
- Machine play
When the game starts, it will ask to select mode.
Pressing S will select the Manual mode.
In manual mode, you have to decide when to wait and when to start moving to get the food.
Pressing E will make the ball move to get the closest food from it's current location.
Maintaining the balance between food and temperature is the way to win.
When the game starts, it will ask to select mode.
Pressing M will select the Manual mode.
The model is trained enough to know the conditions of the game. Tweaking the code will break the model since it is not dynamic.
Please see the jupyter notebook where we analyze the data. This includes the machine play data which is exported to a csv. We then use Matplotlib to plot temperature against energy. We noted a clear spacial distinction between the data points which resulted in the decision to either move or stay still.
The plot can be drawn by running the plot_data jupyter notebook file.