performance - Comparing two neural networks (nntool in Matlab) -

i'm new neural network toolbox (nntool) in matlab. have trained 2 networks using same data set. 1 of these networks contains higher number of neurons other one.

now i'm wondering: how can compare these networks? how can network better network b?

is number of correctly classified pattern in test set? lets both networks shown same test set , network classified more pattern correctly. can network (in general) better network b?

or should @ performance according performance function?

are there other measures comparing 2 networks trained different parameter?

that depends on concern. see, in cases analyzing predicted labels, or accuracy of nets can lead pickup decision, when networks have shallow architectures,however there side-handed issues may become more important when decide see nets wider eyes.

• for example, in training phase, adding 1 hidden unit first hidden layer comes inserting d (dimension of input layer) free parameters (weights) model should estimated. in other hand, more free parameters model has, more training data required come reliable model. therefore, bigger networks well-accepted long have enough data compensate added free parameters. rule of thumb, inserting more free parameters increase chance of over-fitting has been vital problem in deep neural networks , many efforts has been made resolve it.
• another case less important in shallow nets, computational cost imposed hidden nodes. since looking wide eyes, mentioning issue necessary. in cases when network goes deeper, computational cost becomes more challenging. computational cost in training phase important issue when use back-propagation update parameters.
• one other thing may see in deep neural networks memory requirements. number of layers or neurons increase, number of free parameters grows dramatically such in deep networks may see millions of parameters. clear loading amount of parameters asks sufficient hardware requirements.

hope helps.