How to see all Data Types in Performance Chart after training a network in nntraintool

1 view (last 30 days)
Hello everyone,
I've just trained a Network on a Data set, but I'm not able to see all three Data types in there. It shows just one curve labeled for test data and a green circle(for validation?!), shown in the Picture below (however the convergence behavior seems to be for Train data, I guess).
I've checked the way I've set up the Data and its form for the network. but it's OK. I've run the samples in Deep learning toolbox documentation( Cancer Detection, Wine Classification and Crab Classification). They've performed well, as written in Documentation, and all three curves were shown. I can't detect what's wrong in my case.
Thanks for your help!

Answers (1)

Walter Roberson
Walter Roberson on 22 May 2020
The missing lines are "underneath" the red line. The red line was drawn last, so it is on top.
If you look carefully at the corner near epoch 5 and MSE .6 or so, you can see a place where the green is showing through where the match was not identical. See also from about 525 to 600 the green shows slightly. The blue is underneath the red, it appears to me.
If you zoom in you can probably separate the green from the red better.
  2 Comments
SayedPedram Hosseini
SayedPedram Hosseini on 22 May 2020
It's right and is excatly the problem, that the lines are so close to each other. I've changed the number of hidden layers of Network and the training function many times, but always all curves are on each other.
I think this result is weird and couldnt be a correct training of a network, though the Data set has a large number of samples.
Greg Heath
Greg Heath on 25 May 2020
This is a common result for good designs.
One hidden layer is sufficient for any problem.
Minimize the number of hidden nodes to mitigate overtraining an overfit net
Hope this helps.
THANK YOU FOR FORMALLY ACCEPTING MY ANSWER
Greg

Sign in to comment.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!