Clarification needed for array-wise input to network training

Hello Comunity,

I have been trying to solve some problems with both feedforward and NARX networks, although I need to admit I am pretty new to the whole topic and I hope you can help me with the following:

Just a few minutes ago I tried the graphical net generation from the toolbox again and I was surprised to see the following:

Inputs 'Inputs' is a 1x200 cell array of 6x3966 matrices, representing dynamic data: 3966 series of 200 timesteps of 6 elements.

So far I thought that each cell represents one batch. The rows of the matrices within the cells represent the different Inputs and the columns are the time steps. That was at least how I understood it from the documentation.

Now my Dataset looks like the following:

200 Matrices correspond to 200 different tests aprox. 2000 Timesteps each 6 physical messurable values per timestep.

I know that the time resolution is extremely high, but that is parameter I am not supposed to change that, due to the high dynamic rate of my dataset.

What I wanted to do is to seperate the Matrices in different batches for which the timesteps will be split. Therefore it would be nice to know which format is the right one in this case.

Either

Cell -> batch size number of Rows -> Input nodes number of columns -> Timesteps

or something like this:

Cell -> Timesteps number of Rows -> Input nodes number of columns -> ?series?

Thank you very much for your help.

Kind regards

Dominic