Simulate default patternnet with feedforwardnet in Matlab?

Question

I got very different training efficiency with the following network

net = patternnet(hiddenLayerSize);

and the following one

net = feedforwardnet(hiddenLayerSize, 'trainscg');
net.layers{1}.transferFcn = 'tansig';
net.layers{2}.transferFcn = 'softmax';
net.performFcn = 'crossentropy';

on the same data.

I was thinking networks should be the same.

What thing I forgot?

UPDATE

The code below demonstrates, that network behavior is uniquely depends on network creation function.

Each type of network was ran two times. This excludes random generator issues or something. Data is the same.

hiddenLayerSize = 10;

% pass 1, with patternnet
net = patternnet(hiddenLayerSize);

net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

[net,tr] = train(net,x,t);

y = net(x);
performance = perform(net,t,y);

fprintf('pass 1, patternnet, performance: %f\n', performance);
fprintf('num_epochs: %d, stop: %s\n', tr.num_epochs, tr.stop);

% pass 2, with feedforwardnet
net = feedforwardnet(hiddenLayerSize, 'trainscg');
net.layers{1}.transferFcn = 'tansig';
net.layers{2}.transferFcn = 'softmax';
net.performFcn = 'crossentropy';

net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

[net,tr] = train(net,x,t);

y = net(x);
performance = perform(net,t,y);

fprintf('pass 2, feedforwardnet, performance: %f\n', performance);
fprintf('num_epochs: %d, stop: %s\n', tr.num_epochs, tr.stop);

% pass 1, with patternnet
net = patternnet(hiddenLayerSize);

net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

[net,tr] = train(net,x,t);

y = net(x);
performance = perform(net,t,y);

fprintf('pass 3, patternnet, performance: %f\n', performance);
fprintf('num_epochs: %d, stop: %s\n', tr.num_epochs, tr.stop);

% pass 2, with feedforwardnet
net = feedforwardnet(hiddenLayerSize, 'trainscg');
net.layers{1}.transferFcn = 'tansig';
net.layers{2}.transferFcn = 'softmax';
net.performFcn = 'crossentropy';

net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

[net,tr] = train(net,x,t);

y = net(x);
performance = perform(net,t,y);

fprintf('pass 4, feedforwardnet, performance: %f\n', performance);
fprintf('num_epochs: %d, stop: %s\n', tr.num_epochs, tr.stop);

Output follows:

pass 1, patternnet, performance: 0.116445
num_epochs: 353, stop: Validation stop.
pass 2, feedforwardnet, performance: 0.693561
num_epochs: 260, stop: Validation stop.
pass 3, patternnet, performance: 0.116445
num_epochs: 353, stop: Validation stop.
pass 4, feedforwardnet, performance: 0.693561
num_epochs: 260, stop: Validation stop.

Answer 1

Looks like those two aren't quite the same:

>> net = patternnet(hiddenLayerSize);
>> net2 = feedforwardnet(hiddenLayerSize,'trainscg');
>> net.outputs{2}.processParams{2}

ans =

    ymin: 0
    ymax: 1

>> net2.outputs{2}.processParams{2}

ans =

    ymin: -1
    ymax: 1

The net.outputs{2}.processFcns{2} is mapminmax so I gather that one of these is re-scaling it's output to match the output range of your real data better.

For future reference, you can do nasty dirty things like compare the interior data structures by casting to struct. So I did something like

n = struct(net); n2 = struct(net2);
for fn=fieldnames(n)';
  if(~isequaln(n.(fn{1}),n2.(fn{1})))
    fprintf('fields %s differ\n', fn{1});
  end
end

to help pinpoint the differences.