How to get deep learning layer output size?

Hi,

I am getting to know MATLAB's capability with deep learning (I am fluent in TensorFlow).

I built a very big computation graph.

My question is :

How can I get a specific layer's output sizes without training the network? (for dubugging purposes)

I know that I can do:

features = activations(net,X,layer)

but this requires me to train the network (I want to avoid it since it will take a while).

Thanks

Answer 1

[Maria Duarte Rosa]

Hi Itzik,

Have you tried analyzeNetwork?

https://www.mathworks.com/help/deeplearning/ref/analyzenetwork.html

[Itzik Ben Shabat]

Thanks. It seems to be just what I was looking for.

[Mirko Prezioso]

Hi, is there a way to extract the data from AnalyzeNetwork? I mean not just having them in graphical form but as output arrays to be able to use the data. I am looking for the Learnables total size.

Thanks

[Utkarsh Virtuous]

net.Layers().Weights give you the learned parameters. Specify the layer number in bracket.

[Angelo Yeo]

@Mirko Prezioso: A new class NetworkAnalysis will be added from R2024a. This class can be retrieved as an output of analyzeNetwork function, and will capture the output of analyzeNetwork. This will include a property called "LayerInfo" as a table.