MATLAB Answers

## how to predict from a trained neural network ?

Asked by farfar

### farfar (view profile)

on 15 Jan 2018
Latest activity Commented on by Greg Heath

### Greg Heath (view profile)

on 15 Feb 2019
Accepted Answer by Greg Heath

### Greg Heath (view profile)

Hello I am trying to use neural network to make some prediction based on my input and target data. I have read all related tutorial in Matlab and also looked at the matlab examples. I kinda learned how to develop a network but I dont know how to use this train network to make some prediction ? is there any code that im missing ? does anyone have a sample script that can be shared here?
that's what I have, for example : x=[1 2 3;4 5 3] t=[0.5 0.6 0.7] , net=feedforwardnet(10) , net=train(net,x,t) , perf=perform(net,y,t)
how can I predict the output for a new set of x (xprime=[4 2 3;4 7 8]) based on this trained network? thanks

#### 0 Comments

Sign in to comment.

## 2 Answers

Answer by Greg Heath

### Greg Heath (view profile)

on 16 Jan 2018
Accepted Answer

1. Your code should yield an error because you have not defined y.
here are two ways to define output y, error e and normalized mean square error NMSE (= 1-Rsquare)
1. [ net tr ] = train(net,x,t);
y = net(x);
e = t-y;
2. [ net tr y e ] = train(net,x,t); % My favorite
then, in general,
NMSE = mse(e)/mean(var(t',1))
or for 1-dimensional outputs
NMSE = mse(e)/var(t,1)
Hope this helps.
Greg

#### 0 Comments

Sign in to comment.

Answer by Mritula C

### Mritula C (view profile)

on 14 Feb 2019

Hi How do you predicted with a new test class?

Greg Heath

### Greg Heath (view profile)

on 15 Feb 2019
1. You misplaced your commented question into an Answer Box.
2. This is a regression prolem. Your question involves classification.
Greg

Sign in to comment.