itis-enrico-fermi / planarevolvedantenna Goto Github PK

The impact of a crossover is not clear. Crossovers should move the candidate solution farther in space, while a single mutation should help exploring the neighbourhood. This translates in crossover generating solution that have a higher Hamming distance than those generated by single mutations, but whenever part of the rod changes, all the pieces after change with it.

Should we move to a first degree spline representation to reduce the effect of a mutation on the solution?

Trend of:

• standard deviation
• mean fitness
• max fitness
• convergence speed
• killed genes (we should notice a decreasing trend after #3 will be closed)

Set a default configuration and review the configuration file loading and parsing.
Move configuration inside the core module, if used.

Electromagnetic analysis on genes with intersecting segments generate an error.
This is a problem on first epochs where no gene might have a positive fitness, thus making the parent choice fail.
We could fall back to a fully random extraction for parents if roulette method fails because the weights are non-positive.

Deriva cromosomica

1. Estrazione individuo random
2. Selezione del suo intorno
3. Estrazione individui con probabilità proporzionale alla fitness
4. Sostituzione individuo più debole