Release 0.5

There now is the possibility to safe your networks and load them back in.

Currently it is only applied to the WeightedNetwork, which is also the most used network. It is being saved to a JSON-String, since this is far better than creating a custom save-format or using XML.

Also, when using the PropagateAllNetworks-Method, all of the Networks are now being propagated with Tasks for faster processing.
The Breeding-Process is still single-threaded, because the Random-Class is not Thread-safe and locking the built-in Random-Generator may result in performance-issues which would make it useless to breed the networks async anyway.

This update requires you to include the Newtonsoft-DLL delivered in this package

Release 0.4

Replaced Dictionary with List<KeyValuePair<IWeightedNetwork, double>> because Dictionary is implementing a Hash-Table that sometimes reorders the keys, so the Networks, so that when trying to access the networks they could be returned in random order. This is prevented now with using List.

Fixed Inbuilt Fitness-Calculation so that the Expected is now subtracted from the output instead of added, which stops making negative values good where positive numbers are wished (-0.99999 was as good as 0.99999 even though 1 was wished)

Fixed FitnessBasedPoolGenerator so that Networks with 0 fitness are ignored to stop infinite loops when trying to multiplicate until bigger than 1

Version 0.3.1 - Bugfixes

Quick Bigfixed

• Fixed a bug where when FitnessBasedPoolGenerator was used together with RandomNetworks, that it then caused a infinite look because it was constantly trying to multiply the fitness of 0 to a value of over 1. This was caused by adding the new RandomNetworks before creating the Breeding-Pool, and adding the RandomNetworks would give the networks a fitness of 0.
• Fixed wrong Values being assigned at the NewNetwork-Method

Version 0.3

Added a Random-Network-Amount to GeneticAlgorithm that controls the amount of totally randomly created Networks each Breeding to keep better control over Variety in the network
Added the possibility to switch the PoolGeneration-Method used by assigning another PoolGenerator to the Genetic Algorithm

• Currently there are the IndexBased- and the FitnessBased- Pool-Generators

Version 0.2

• Made it optional to pass ExpectedData when Setting up a Test
• Made it possible to set the Fitness of all Networks or individual Networks.

First Release

First Release, where everything actually is already implemented, Bu I dont wanna call it 1.0 for now