... | ... | @@ -110,6 +110,22 @@ During the genetic call so called ['gods'](https://git.opendfki.de/reuschling/ge |
|
|
genIe.addEvolutionGod(GeneticParamOptimizerGod evolutionGod)
|
|
|
```
|
|
|
|
|
|
## Candidate vector metadata
|
|
|
|
|
|
A fitness function can attach metadata to a candidate vector, together with it's calculated fitness. GenIe doesn't calculate anything with this metadata, it just gives parent population metadata to siblings, where the fitness function has the possibility to do something special with it.
|
|
|
|
|
|
This is more or less a communication metaphor, where the former generation can give dedicated information to it's siblings. One example would be e.g. the file path to a trained neural net model from a parent, where a sibling then has the chance to train further on top of this pre-trained model. (Population based training)
|
|
|
|
|
|
A fitness function receives parent generation information as follows:
|
|
|
1. The metadata from the direct parents (GenIe generates a sibling out of two parents)
|
|
|
2. The metadata from the best top N elite candidate vectors of the whole parent generation
|
|
|
3. The metadata from a randomly picked entry of the elite set from point 2. (Can also be empty as valid random pick)
|
|
|
|
|
|
|
|
|
Remark:
|
|
|
* The first generation of candidate vectors doesn't have any parents. Thus you can specify the parents metadata for these vectors manually in the config file
|
|
|
|
|
|
|
|
|
## Result interpretation with entropy analysis
|
|
|
|
|
|
During optimization the software gives further information about the parameters that are optimized. This comprises hints how big the impact of a single parameter to the final search result quality is. It gives you an overview of the relevancies of the parameters, for e.g. deciding to remove unnecessary parameters at all.
|
... | ... | |