6244RE: Starting with Neat, the XOR problem, some questions
- Jan 17, 2014Hi Simon, it's true that it's difficult to go through this logic in detail on the group. You might want to email me offline if my answers here don't help.
Some of what you're saying doesn't quite sound right to me, but I'm not entirely sure given that it's not program code. But there indeed may be some bugs there. Regarding connection and node genes, when you do crossover, all the connection genes and all the node genes should be taken from the parents. That is, no new numbers are generated during crossover. Only connection or node adding mutations can introduce new numbers.
Also, not all reproduction has to be through mating. You can have a single parent that produces offspring. So if someone is alone in a species, they can have children on their own. I didn't fully understand your questions about one species needing to mate with another. There is never a situation where one species absolutely must mate with another. You can allow it to happen sometimes at a low rate if you want to, but nothing necessitates it.
The population size also should never change.
Regarding your question on speciation, you do not need to check every individual against every other individual in the original NEAT speciation algorithm (others have since introduced variant speciation algorithms but I'm speaking of the original one I wrote). In that approach, you only have to compare individuals to species representatives. Each species has a representative and you are comparing each individual in the loop to all those representatives. If the individual is not compatible with any representative, then it becomes the representative of a new species.
- << Previous post in topic