Dimensions of T and Y in a customized regression layer

I did not understand what are the dimensions of T and Y in a customized regression layer (reference : https://it.mathworks.com/help/deeplearning/ug/define-custom-regression-output-layer.html). Looking at my specific case, I have a training dataset with 28x70000 observed input sequences and 1x70000 observed target values. In my case, I am trying to predict the full target sequence. This custom regression layer is the last layer of the network. Before this custom regression layer, I have a fullyConnectedLayer that outputs a scalar (I don't know if this information could be useful in some way). Anyway, what do T (target) and Y (observations) represent? What are their dimensions in my specific case? For instance, I would have expected T to have a size of 1x70000, but I am sure this is not true because I also tried to transpose T and it gave an error, hence it can be probably a multidimensional Matrix Thank you very much!

Answer 1

[Cris LaPierre]

Comments in the code example you link to provide the following information

% Inputs:

% layer - Output layer

% Y – Predictions made by network

% T – Training targets

In otherwords, T is your labeled data (one per observation), and Y is the label predicted by your model (one per observation).

It depends what your input layer is, but I suspect the input size should be 28xN, where N is the number of observations in the mini-batch.