differences between net = newff(P,T,​S,TF,BTF,B​LF,PF,IPF,​OPF,DDF) and net = newff(PR,[S1 S2...SNl],{TF1 TF2...TFNl},BTF,BLF,PF) ?

6 次查看（过去 30 天）
Fereshteh....
Fereshteh.... 2014-11-21
评论： Fereshteh.... 2014-12-4
I thought there is no differences between them and just that in the first one we put target at first and in the second one we put size of layers at first, but i wrote a program and when i used this net = newff(P,T,S,TF,BTF,BLF,PF,IPF,OPF,DDF) my neural network stopped working due to "validation checks" at iteration 9 of 1000 epochs but when i used net = newff(PR,[S1 S2...SNl],{TF1 TF2...TFNl},BTF,BLF,PF) it reached to maximum epochs which i set on 1000.
what is wrong?

请先登录，再回答此问题。

采纳的回答

Greg Heath
Greg Heath 2014-11-23
I compared the two formats and obtained the exact same answer on version 2014a.
1. Did you run both nets using the same version? 2. Are you sure the random number generator is initialized to the same state so that the random data division and random initial weights are the same?
Hope this helps
*Thank you for formally accepting my answer.
Greg
  1 个评论
Fereshteh....
Fereshteh.... 2014-12-4
hi Greg thanks for your answer. i use version 2013b. I don't get any same answers, for example i wrote these codes:
if true
[ x, t] = iris_dataset;
xl=1:1:150;
net=newff(xl,x,90);
net.trainparam.epochs=1000;
net=train(net,xl,x);
Y=sim(net,xl);
el=mse(x-Y)
NMSE=el/mean(var(x',1));
net=newff(minmax(xl),[90,4],{'tansig' 'purelin'},'traingd');
net.trainparam.epochs=1000;
net=train(net,xl,x);
Y=sim(net,xl);
el=mse(x-Y)
NMSE=el/mean(var(x',1));
end
i used x in iris_dataset as my target and i used xl=1:1:150 as my inputs! i don't get same result or any close results. i do know that the random initial weights are not the same but i expect the results not to be so different.
i also wonder in newff(p,t,s) performance (plotperform) we see all training -validation and test curves but in newff(minmax(p),s) performance (plotperfom) we just see training curve! i wonder if it means that in the newff(p,t,s) matlab separates data into 3 groups of training and validation and test by default but in the latter it doesn't do it.
as i see there are many differences between these 2, i do need help to understand where i am being wrong !
regards Fereshteh

请先登录，再进行评论。

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

Community Treasure Hunt

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

Start Hunting!

Translated by