prune

Delete neural inputs, layers, and outputs with sizes of zero

Syntax

[net,pi,pl,po] = prune(net)


Description

This function removes zero-sized inputs, layers, and outputs from a network. This leaves a network which may have fewer inputs and outputs, but which implements the same operations, as zero-sized inputs and outputs do not convey any information.

One use for this simplification is to prepare a network with zero sized subobjects for Simulink^®, where zero sized signals are not supported.

The companion function prunedata can prune data to remain consistent with the transformed network.

[net,pi,pl,po] = prune(net) takes a neural network and returns

net	The same network with zero-sized subobjects removed
pi	Indices of pruned inputs
pl	Indices of pruned layers
po	Indices of pruned outputs

Examples

Here a NARX dynamic network is created which has one external input and a second input which feeds back from the output.

net = narxnet(20);
view(net)

The network is then trained on a single random time-series problem with 50 timesteps. The external input happens to have no elements.

X = nndata(0,1,50);
T = nndata(1,1,50);
[Xs,Xi,Ai,Ts] = preparets(net,X,{},T);
net = train(net,Xs,Ts);

The network and data are then pruned before generating a Simulink diagram and initializing its input and layer states.

[net2,pi,pl,po] = prune(net);
view(net2)
[Xs2,Xi2,Ai2,Ts2] = prunedata(net,pi,pl,po,Xs,Xi,Ai,Ts);
[sysName,netName] = gensim(net2);
setsiminit(sysName,netName,net2,Xi2,Ai2);

Version History

Introduced in R2010b