Train a Neural Network with multidimensional matrices

I want use neural network to perform binary prediction for a set of 10 robotic manipulators that have to converge to a target position when arranged in a crowded environment. Each robot is characterized by 5 feature and my dataset is composed of 2000 simulations.Then if I indicate with N the number of features of the robots, with M the number of simulations and with R the number of the robots, I have a N x M x R matrix. The problem is that i cannot feed a Neural network with the data in this form and I am looking for another way to arrange the data for the Network. It is important that when i have to split the data for training, validation and test, the split is performed on the number of simulation M, and not on the number of robots R.

Thank you

Answer 1

[Srivardhan Gadila]

If you are thinking of a fullyconnected network then use the imageInputLayer for giving the input to the network and use the dividerand for splitting the dataset across M. Reshape your N x M x R matrix to N x R x 1 x M or R x N x 1 x M matrix using reshape function

Also refer to List of Deep Learning Layers.