ludvigolsen/evolutionarycomputing issues and pull requests

#16 - Implement self-adaptation

Issue - State: closed - Opened by LudvigOlsen over 6 years ago - 1 comment

#15 - Add simulated binary crossover. Find formula online.

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#14 - n-crossover recombination with rolling, to have less positional bias.

Issue - State: open - Opened by LudvigOlsen over 6 years ago - 1 comment

#13 - Add extra genotypes, that decides the recombinator and mutator. This way, the evolution will choose the best methods. (See self-adaptation)

Issue - State: open - Opened by LudvigOlsen over 6 years ago

#12 - When having multiple recombinators, make the breeder choose randomly between them.

Issue - State: open - Opened by LudvigOlsen over 6 years ago

#11 - All genotypes in genome 3 (seed 3) goes towards 0 in KatsuuraEvaluation

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#10 - Monitor avg. fitness score during training.

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#9 - Changing representation during the simulation. E.g. Start with integer representation and then when the fitness curve flattens, change to float representation for finetuning. (Would probably be float/double entire time, just always rounded)

Issue - State: open - Opened by LudvigOlsen over 6 years ago

#8 - Their evaluate function sometimes returns null. Figure out why and how often.

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#7 - Max score doesn't change throughout the simulation

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#6 - Now that individuals have a unique ID, we can track their family in a tree, to make sure incest doesn't happen. This way, they can continue to be alive, just not have children with their own children.

Issue - State: open - Opened by LudvigOlsen over 6 years ago

#5 - Implement multiple "tribes". I.e. multiple populations that only mix every 1000 rounds, or similar. All tribes have different initialization, methods for recombination and mutation, and perhaps a exploitation vs exploration rate.

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#4 - SphereFunction change from negative to positive total fitness. Leads to negative probabilities. Scale the fitness function somehow?

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#3 - Idea: Implement neural nets in individuals

Issue - State: open - Opened by LudvigOlsen over 6 years ago

#2 - Write Population class

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

#1 - Write Child class

Issue - State: closed - Opened by LudvigOlsen over 6 years ago

Ecosyste.ms: Issues

GitHub / ludvigolsen/evolutionarycomputing issues and pull requests

#16 - Implement self-adaptation

#15 - Add simulated binary crossover. Find formula online.

#14 - n-crossover recombination with rolling, to have less positional bias.

#13 - Add extra genotypes, that decides the recombinator and mutator. This way, the evolution will choose the best methods. (See self-adaptation)

#12 - When having multiple recombinators, make the breeder choose randomly between them.

#11 - All genotypes in genome 3 (seed 3) goes towards 0 in KatsuuraEvaluation

#10 - Monitor avg. fitness score during training.

#9 - Changing representation during the simulation. E.g. Start with integer representation and then when the fitness curve flattens, change to float representation for finetuning. (Would probably be float/double entire time, just always rounded)

#8 - Their evaluate function sometimes returns null. Figure out why and how often.

#7 - Max score doesn't change throughout the simulation

#6 - Now that individuals have a unique ID, we can track their family in a tree, to make sure incest doesn't happen. This way, they can continue to be alive, just not have children with their own children.

#5 - Implement multiple "tribes". I.e. multiple populations that only mix every 1000 rounds, or similar. All tribes have different initialization, methods for recombination and mutation, and perhaps a exploitation vs exploration rate.

#4 - SphereFunction change from negative to positive total fitness. Leads to negative probabilities. Scale the fitness function somehow?

#3 - Idea: Implement neural nets in individuals

#2 - Write Population class

#1 - Write Child class