Ideas

Bugs:
- Elite saving goes nan
(- when retraining an already trained generation, result is not reproducible by starting from beginning)

Ideas:
- preserve gene net parameters under mutation
- cluster center in exploration
- population score / auto end
(?) restructure genomes for distance plot / dist to species repr
(-) inter-gene crossover?
- Every gene has a detail to its weight. e+v of all weights, e+v in regions (e.g. col,row,blocks), all weights
  Higher detail -> higher penalty
- Mutation chances change dynamically (What works gets boosted)
(-) dilation, bias for Kernel
(-) Mutation exception/variance
    Same for the selection of what genes to mutate
- stride dial for Pooling
(-) better structure separate algorithm / initialization/ data
- Performance/ Score einstellbar
- Save/Load more useful

Paper:
- Analyse von learning rate etc / macht Evolution das richtige

Todo:
- Viele Testläufe um gute Werte zu finden für:
   - Elitism
   - Selection
   - ...