I want to implement Multilayer perceptron for software effort estimation. Which function should I use feedforwardnet, fitnet or something else.
1 view (last 30 days)
Show older comments
sushma khatri
on 6 Aug 2019
Commented: sushma khatri
on 12 Aug 2019
My input may be continuous or categorial. but target is always a continuous no.
0 Comments
Accepted Answer
Maria Duarte Rosa
on 9 Aug 2019
For continuous outputs both fitnet and feedforwardnet are equivalent and the natural choice.
Perhaps an easier way to get strated is using the app:
If further customization is needed then 'network' allows one to build more flexible networks:
Note: for deep learning networks a good way to get sarted is by using the Deep Network Designer app:
I hope this helps.
3 Comments
Maria Duarte Rosa
on 12 Aug 2019
Hi Sushma,
The targets for feedforwardnet are continuous. Please see here:
[x,t] = simplefit_dataset;
net = feedforwardnet(10);
net = train(net,x,t);
t are continuous.
Petternnet is for categorical targets, see here:
[x,t] = iris_dataset;
net = patternnet(10);
net = train(net,x,t);
t in this case is the 1/0 form.
More Answers (1)
sushma khatri
on 12 Aug 2019
2 Comments
Maria Duarte Rosa
on 12 Aug 2019
fitnet and feedforwardnet are equivalent. You can use one or the other. If 'effort' is always continuous then fitnet or feedforwardnet seems to me to be the most natural choice for your task.
nftool is the most appropriate UI for modelling continous outputs.
nprtool is for patternnet where the output is categorical.
ntstool is for modelling time-series, using narxnet and other similar networks.
See Also
Categories
Find more on Modeling and Prediction with NARX and Time-Delay Networks 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!