2
$\begingroup$

What does the Regression Plot in the Matlab Neural Network Toolbox show? I thought I understood it when I looked at a univariate regression plot, but I've just plotted one for multivariate regression, and it makes no sense to me.

My Neural Network takes in 24 inputs, and gives 3 outputs. The 24 inputs are 24 different angles, and the 3 outputs are the (x,y,z) positions in 3d space. I expect that the function mapping the two is highly nonlinear, but I'm not sure.

Here is the regression plot, and here is the performance plot if that helps.

What does the regression plot show? I do not understand this graph at all. Surely it cant plot the function because that would require a plot in 27 dimensions (24 inputs + 3 outputs). What do the 4 graphs represent? To me it looks like it is saying that the function is linear, could this be true?

$\endgroup$

1 Answer 1

1
$\begingroup$

These plots give you an idea of how close the output from your model is to the actual target values. My impression looking at them is that the plotted values are similar to the residuals from regular linear regression.

Matlab provides some discussion on how to interpret these plots on this page. If you CTRL+F "regression plot" you will find the relevant section.

Looking at your results, it seems you have a pretty decent fit, though you seem to have some outliers you might want to look into.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.