An experiment with genetic algorithms to draw an image starting from the bottom-left corner, without skipping any pixels (i.e. without lifting the pencil as an analogy). The results with the default parameters with a dynamic mutation rate are shown in the table below.
| Original | Evolution | Result |
|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Different approaches were tried, as explained below. Individuals in this context are basically paths (up, down, right, left) for the "pencil" to follow starting from the bottom-left corner.
- Individuals are created randomly and then sorted according to their
fitness. - The top-k individuals are chosen as parents and crossover is applied to them to complete the population.
- Mutation is applied to the newly created individuals, and those mutants are passed with their parents to the next generation.
- Individuals are created randomly and then sorted according to their
fitness. Fitnessvalues are accepted as weights andkparent are chosen according to a weighted probability from the population.- Crossover is applied to the parents and the population is completed using mutants.
- The parents and the newly created mutants are passed on to the next generation.
The mutation rate can be decreased linearly. It reaches 0.01 after half the generations are passed.
The path length can be increased starting from the diagonal length up to the path length set in the code.
Results of different experiments with different methods and hyperparameters are shown in the image below.
