Problem using scalingLayer for shifting actor outputs to desired range

Greetings everyone. I want to use scalingLayer to shift my actor network's outputs to desired range. I'm using this reference and the code below :

numObs = 10;
numAct = 5;
actorLayerSizes = [400 300];
sc = reshape([10, 10, 20, 15, 15], [1 1 5]);
bias = reshape([0, 0, 0, 0, 0], [1 1 5]);

actorNetwork = [
    imageInputLayer([numObs 1 1],'Normalization','none','Name','observation')
    fullyConnectedLayer(actorLayerSizes(1), 'Name', 'ActorFC1')
    reluLayer('Name', 'ActorRelu1')
    fullyConnectedLayer(actorLayerSizes(2), 'Name', 'ActorFC2')
    reluLayer('Name', 'ActorRelu2')
    fullyConnectedLayer(numAct, 'Name', 'ActorFC3')                       
    tanhLayer('Name','ActorTanh1'), ...
    scalingLayer('Scale',sc, 'Bias',bias)
    ];

I'm getting the error below:

Error using vertcat

Dimensions of arrays being concatenated are not consistent.

Can anyone help me with this?

Thanks

[Emmanouil Tzorakoleftherakis]

Which agent are you using? Depending on the agent, the actor architecture may change.

[Mostafa Nazmi]

I'm using DDPG agent for this problem

[Mostafa Nazmi]

@Emmanouil Tzorakoleftherakis Any ideas how to fix this my friend?

Answer 1

[Emmanouil Tzorakoleftherakis]

If you remove ',...' from the 'tanh' row the error goes away. The way you have it now, you are adding the scaling layer in the same row as tanh whereas it should be in its own row.

actorNetwork = [
    imageInputLayer([numObs 1 1],'Normalization','none','Name','observation')
    fullyConnectedLayer(actorLayerSizes(1), 'Name', 'ActorFC1')
    reluLayer('Name', 'ActorRelu1')
    fullyConnectedLayer(actorLayerSizes(2), 'Name', 'ActorFC2')
    reluLayer('Name', 'ActorRelu2')
    fullyConnectedLayer(numAct, 'Name', 'ActorFC3')                       
    tanhLayer('Name','ActorTanh1')
    scalingLayer('Scale',sc, 'Bias',bias)
    ];

[Mostafa Nazmi]

Thanks a lot my friend

[Emmanouil Tzorakoleftherakis]

Np. Please accept the answer if it answered your question

[Mostafa Nazmi]

All done and thanks again!