Manna Mouse:
Exploratory Fitness Landscaping

A joint project of and Caplet™Security

Among interactive artificial life simulations, Manna Mouse is uniquely simple: a creature's genome encodes its location on the display area. This simplicity allows us to see interesting dynamics that are often inaccessible.

With your mouse you paint manna--the reward function. By visualizing the fitness landscape, you can quickly build intuitions about evolution.

To start, press go. how to play
changing parameters
source code

Please enable Java in your browser.

Manna Mouse Applet

You see the evolution of two, separate populations of creatures. A creature's genotype encodes two values--an x coordinate and a y coordinate. The phenotype is expressed as a dot on the display area.

For each population, the first generation is determined randomly. Without a fitness function to selectively steer reproduction, subsequent generations of creatures appear to be moving randomly through the 2-d space.

Genetic Algorithms were invented by John Holland. See my book review of John Holland's Hidden Order, appearing in the December 1996 issue of Reason Magazine.

Painting Manna

Click stop.

Mouse down in one of the population areas. Slowly, drag the mouse to paint manna. Start with something simple, such as a vertical line. The manna appears in both populations.

Click go.

The distribution of manna is the fitness function; creatures located on manna are selected for, during reproduction. But manna is consumed, and eventually, the population returns to random movement.

For GA aficionados, we are using tournaments rather than roulette wheels, and we do collision avoidance: if a mating would result in creating a creature in an occupied location, we try again rather than redundantly search.

Manna Mouse Controls

  • go: Start or resume the simulation. You can change parameters and draw manna while the simulation is running. To stop it, click stop.
  • stop: Stop the simulation. To resume it, click go.

Manna Mouse Parameters

Note: If you do not see the ui controls for changing the algorithm's parameters scroll the applet into view in the browser window.

Set the parameters the same in both populations and watch a few runs; you will be surprised at the divergence. This is known as sensitivity to initial conditions: where an evolutionary process ends up depends not only on the rewards but where it starts.

  • population: The population size can be changed. When increased, the new creatures are placed in randomly-determined locations.
  • mutants: The mutation rate defines the percentage of the evolving population that has a random bit flipped. These random mutations can increase the genetic diversity over what is seen with crossover alone.
  • gray: A Gray representation encodes the genotype in a way that results in adjacent positions differing only by one bit. Binary representations do not have this property. If a transformation happens to invoke the hardware's carry operation, adjacent positions will have very dissimilar genotypes.

    Note: Gray representation has not been implemented yet -- the default is binary.
  • uniform: When two creatures mate, how are their genotypes combined to create their offspring? The two options are uniform crossover and two-point crossover (seen in bacteria).

    In uniform crossover the contribution from each parent is determined randomly.

    In two-point crossover, the two ends of the genotype are joined to form a ring, removing the significance of the end points. The ring is cut at two positions, determined randomly. The parents contribute alternate segments to their offspring. For example, if the rings are cut at positions A and B, the child's genotype is made up of the segment from A to B of one parent and the segment from B to A of the other parent.
  • local: Are creatures more likely to mate if they are closer together?

    Note: Spatial locality has not been implemented yet.
  • consumed: Is manna consumed when it is eaten? The pure notion of a fitness landscape (or function optimization) would not allow for manna consumption, but it makes for more interesting dynamics. As the manna is consumed, its color changes from green to red before finally disappearing.

Javadoc-umentation, and source code (AWT 1.0):