How do i approximate non-linear function with neural networks and work with performance plot?

I have to approximate nonlinear function with neural network. The number of layers and number of neurons in layers are chosen by me. Suppose i have a certain dataset (attached).

Dataset contains 7 variables. T1-T5 are used to create function (see code) and X1 and X2 are inputs.

Now my code just does not work as intended. I am not even sure this is correct so a confirmation would be nice ☺.

But now the real problem. After i train the network, the performance graph shows three lines, one for training, one for validation and one for test. How can i change it. I only used train command for training set and it divided the training set into validation, training and test.

Lets suppose I join those three together into the big one. How does the algorithm decide how to divide them? Does it divide the data evenly? Can i decide how it will be divided?

load 'Data_Problem1_regression.mat';
Tnew = (8*T1 + 5*T2 + 2*T3 + 1*T4 + 0*T5)/(8+5+2+1+0);  %this is given from the exercise. This is the function output
training = Tnew(1:1000)';                                % 1000 samples from Tnew are supposed to be training (transposed for network to work)
validation = Tnew(1001:2000)';                           % 1000 for validation
test = Tnew(2001:3000)';                                 % 1000 for testing.
x = {X1(1:1000)';X2(1:1000)'};                           % Neural netowork will have two inputs.
%%
net1=feedforwardnet(12,'trainlm');
net1.numinputs = 2;
view(net1)
net1 = train(net1,x,training);                          % only giving "training" set as an argument but the performance plot will have 3 lines

Answer 1

[Srivardhan Gadila]

The net1 from above has the function divideFcn as Object property. This property defines the data division function to be used when the network is trained using a supervised algorithm, such as backpropagation. You can set this property to the name of a division function.

For a list of functions, type "help nndivision" in command window

If you don't want any validation or test splits then use dividetrain.