Every learning method now inherits from the abstract class LearningMethod, instead from the interface ILearningMethod.

ENH + FIX: Fixed output problems of MSE evaluation and the data-set is not de-normalized anymore, when evaluating an error, thus the error can be evaluated multiple times (e.g. using different networks).

FIX: Fixed non-normalizing eval() method in ErrorFunctions + rewritten ErrorFunction classes to prevent duplication of code.

ADD: added proggress debug output for particle swarm optimization

FIX: Removing max and min values after de-normalization of data-set, so the is_normalized() method will again return 'False' after de-normalization.

FIX: fixed an issue with NaN in GradientDescent caused by acos(x), added more debug info to errorfunction validation

ENH: Added check when DataSet constructor gets path to the file, which is not a correct DataSet archive.

ENH: Added data check before creating a new DataSet - prevents SEGFAULTs occuring when read() method was not called beforehand.