How can I get a decent number of samples VS train error graph in neural networks?

Hello;

I am trying to obtain the graph for number of samples VS training error. The graph I would like to have is like this

I am trying to do this by using the batch training with 'trainlm' algorithm. The code I have written is the following:

clc, clear all, close all;

for ctr=1:20

% number of samples is ctr

input = linspace(-3,3,ctr);

Target = tanh(input);

% define the net

net=feedforwardnet(1,'trainlm');

% initialize the net

net = init(net);

% parameters are choosen

net.trainParam.epochs = 20;

net.trainParam.max_fail = 6;

% train the net

[net,tr]=train(net,input,Target);

% simulate the net

Y=sim(net,input);

% record the errors and corresponding number of samples

tperf(ctr) = tr.best_tperf;

bestperf(ctr) = tr.best_perf;

end

plot(tperf) xlabel('number of samples'); ylabel('error'); hold on plot(bestperf) legend('tperf','bestperf')

As you can see the error values are saved into an array with corresponding number of samples. However I end up with the graph like that:

It really makes sense that as the number of samples increases the training error will converge to some number, however I cannot do it in my code.

I am looking for your answer.

Yours Sincerely.