Changing the output of a neural network
1 view (last 30 days)
Show older comments
Hi there,
I have a problem with the output of a neural network for data classification. The network has 4 target vectors.
Say for example the output (column vector) is
0.1 0 0.4 0.3
0.1 0.52 0.45 0.25
0.25 0.08 0.1 0.1
0.65 0.4 0.05 0.35
and then I round the output to get a one in each column, and get this
0 0 0 0
0 1 0 0
0 0 0 0
1 0 0 0
and then I use vec2ind(output) to return the rows that contain 1
It returns [4 2].
But columns 3 and 4 and are not classified.
How can I make it return the row with the highest probability estimate? So this example would return [4 2 2 4]
Many thanks
0 Comments
Accepted Answer
Greg Heath
on 5 Feb 2012
>> t = [ 0 0 0 0 % target
0 1 1 0
0 0 0 0
1 0 0 1]
y = [ 0.1 0 0.4 0.3 % output
0.1 0.52 0.45 0.25
0.25 0.08 0.1 0.1
0.65 0.4 0.05 0.35 ]
[ ymax class] = max(y)
t =
0 0 0 0
0 1 1 0
0 0 0 0
1 0 0 1
y =
0.1000 0 0.4000 0.3000
0.1000 0.5200 0.4500 0.2500
0.2500 0.0800 0.1000 0.1000
0.6500 0.4000 0.0500 0.3500
ymax =
0.6500 0.5200 0.4500 0.3500
class =
4 2 2 4
Hope this helps.
Greg
1 Comment
Greg Heath
on 5 Feb 2012
If posterior probability estimates are desired, use the SOFTMAX activation function. However, if LOGSIG is used a reasonable approximation is to normalize the outputs to have a unity sum:
>> yn = y./repmat(sum(y),4,1)
yn =
0.0909 0 0.4000 0.3000
0.0909 0.5200 0.4500 0.2500
0.2273 0.0800 0.1000 0.1000
0.5909 0.4000 0.0500 0.3500
>> sum(yn)
ans =
1.0000 1.0000 1.0000 1.0000
Hope this helps.
Greg
More Answers (0)
See Also
Categories
Find more on Sequence and Numeric Feature Data Workflows in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!