The 0.82 release includes several small, nevertheless very useful new features, some of which I was given the ideas by SNIPE users.
All qualified method names refer to the prefix com.dkriesel.snipe. Changes that may have influence on existing software that uses SNIPE are marked with a .
A new method training.TrainingSampleLesson.splitLesson(double ratio) enables the splitting of an existing TrainingSampleLesson in two by a given ratio. This is useful for creating training- and validation sets of a single training data sets.
A new method training.TrainingSampleLesson.shuffleSamples() shuffles all training samples within a TrainingSampleLesson.
A new method core.NeuralNetwork.setActivation(int neuron, double activation) enables change of the activation values of neurons. This is useful for network topologies like hopfield networks.
The constructor of training.TrainingSampleLesson now clones the arrays of inputs and desired outputs, not just references them.