query about validation set

net = newff(inputs,outputs,14);
net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

what validation type is used in this code like 10 fold cross validation ?

Answer 1

[Greg Heath]

http://www.mathworks.com/matlabcentral/answers/261732-neural-network-time-series-prediction-changing-the-inital-state

Answer 2

[Brendan Hamm]

What is called the validation set in Neural Nets is not quite the same as it is for other machine learning algorithms. Typically a validation set is used to determine how well we believe a fitted model will work (after fitting). In a Neural Net the validation set is used to determine when to stop training the model, that is the validation set should continue to decrease at each iteration but will begin to increase when the model becomes overfit. Therefore the valdiation set is used to determine which iteration provided the "best" result. Finally the testing set is used as comparisson, if it deviates too much from the validation set then the model it might indicate a poor division of these sets.

The actual method of data division is stored in net.divideFcn and by default is randomly sampled.