MC Manual has not been updated in many areas since 2018 (e.g. Kase bars?)
But I am specifically interested in understanding the meaning/ implications of different GA settings.
- when "least fit individuals are discarded ", - how many - as a % of population? is this changeable?
- does a (single) "crossover" always occur between 2 parents ?
Basic GA:
* This algorithm uses non-overlapping generations and Elitism mode (optional).
* For each generation, the algorithm creates an entirely new population of individuals (if the Elitism option is selected, the most fit individuals move
on to the next generation). -- This is not clear at all
-->What exactly happens with the initial population?
-->What are 'non-overlapping generations'?? and how are they created? how is the "entirely new population" created?
-> how is "Crossover probability" applied in this case?
-> how many "Elite/most fit" individual (as a % of population) move to the next generation?
--> if Elitism is not selected, how does Evolution happen altogether?
Incremental GA:
* It simply adds only one or two children to the population each time the next generation is created. These one or two children replace one or two individuals in the previous generation. --> so, does it add or replace?
--> With "Parent" replacement scheme, do children replace parents only if their fitness is better?
- "Number of individuals in population for crossover" --> what's the meaning of this and how does this work?
--> what should it be set to (cpu cores? % of population)?
-> how does "Crossover probability" work in this case?
Convergence Type/Proximal:
* as per manual, GA calculation is stopped after meeting: С [x – N] / C [x] >= P
--> shouldn't it be C[x] / C[x-N] >=P
GA settings
- Polly MultiCharts
- Posts: 264
- Joined: Jul 20 2022
- Has thanked: 1 time
- Been thanked: 67 times
Hello Zheka and shargo,
Basic GA:
The probabilities for Crossover and Mutations are set in Crossover probability and Mutation probability.
When Elitism is turned on, the best are not destroyed (so as not to lose the “elite”).
Non-overlapping generation means that MultiCharts uses one generation for generation.
Incremental GA:
By default, ‘Number of individuals in population for crossover’ is set in the range: from number of CPU cores (so that all cores are loaded during optimization) to Population Size /2. There is not much sense in making this number bigger as it GA will most likely not find the best if the entire generation is updated.
Replacement Scheme is also taken into account (replace the worst/parent/random individual).
- When "least fit individuals are discarded ", - how many - as a % of population? is this changeable?
- does a (single) "crossover" always occur between 2 parents?
Basic GA:
The probabilities for Crossover and Mutations are set in Crossover probability and Mutation probability.
- What exactly happens with the initial population?
- What are 'non-overlapping generations'?? and how are they created? how is the "entirely new population" created?
- how is "Crossover probability" applied in this case?
- how many "Elite/most fit" individual (as a % of population) move to the next generation?
- if Elitism is not selected, how does Evolution happen altogether?[/b]
When Elitism is turned on, the best are not destroyed (so as not to lose the “elite”).
Non-overlapping generation means that MultiCharts uses one generation for generation.
Incremental GA:
- With "Parent" replacement scheme, do children replace parents only if their fitness is better?
- "Number of individuals in population for crossover" --> what's the meaning of this and how does this work?
what should it be set to (cpu cores? % of population)?
By default, ‘Number of individuals in population for crossover’ is set in the range: from number of CPU cores (so that all cores are loaded during optimization) to Population Size /2. There is not much sense in making this number bigger as it GA will most likely not find the best if the entire generation is updated.
Replacement Scheme is also taken into account (replace the worst/parent/random individual).
- how does "Crossover probability" work in this case?
- as per manual, GA calculation is stopped after meeting: С [x - N] / C [x] >= P
shouldn't it be C[x] / C[x-N] >=P