Increasing the number of epochs to reach the performance goal
46 Ansichten (letzte 30 Tage)
Ältere Kommentare anzeigen
dream
am 22 Apr. 2014
Kommentiert: MD Towhid Ur Rahman
am 15 Apr. 2022
Hello,
I am training the neural network with input vector of 85*650 and target vector of 26*650. Here is the list of parameters that I have used
net.trainParam.max_fail = 6;
net.trainParam.min_grad=1e-5;
net.trainParam.show=10;
net.trainParam.lr=0.9;
net.trainParam.epochs=13500;
net.trainParam.goal=0.001;
Number of hidden nodes=76
As you can see ,I have set the number of epochs to 13500. Is it OK to set the number of epochs to such a large number?. Performance goal is not reaching if the number of epochs is decreased and I am getting a bad classification while testing.
1 Kommentar
An Hoang
am 8 Okt. 2017
I thought your epochs is so big and too much training is actually bad because the network overfits to the training data, and then performs badly against new data that it hasn't never seen it before.
Akzeptierte Antwort
Greg Heath
am 22 Apr. 2014
Bearbeitet: Greg Heath
am 2 Mai 2014
[ I N ] = size(input) % [ 85 130 ]
[ O N ] = size(output) % [ 26 130 ]
Ntrn = N - 2*round(0.15*N)% 90
Ntrneq = Ntrn*O % 2340 training equations
%Nw = (I+1)*H+(H_1)*O % unknown weights
% Ntrneq >> Nw, or equivalently, H << Hub where
Hub = -1+ceil((Ntrneq-O) / (I+O+1)) % 20
Try to minimize I and H while using as many defaults as possible.
I don't remember ever having to incease the default numepochs.
Greg
3 Kommentare
Greg Heath
am 23 Apr. 2014
MY APPROACH IS TO VARY NUMBER OF HIDDEN NODES AND TO DESIGN ~ NTRIALS = 10 NETS FOR EACH CANDIDATE VALUE OF H.
I use as many defaults as possible. Rarely have I considered increasing the maximum number of epochs.
MD Towhid Ur Rahman
am 15 Apr. 2022
I like the equations you mentioned here, my question is -"Is there an established paper/ referencce/book that is published by any other author or yourself that metions these equations?" Thanks in advance! If yes, the would you mind sharing the resource with me [Please don't attach links, I just need the Title and Author, I can find it myself]
Weitere Antworten (1)
Greg Heath
am 23 Apr. 2014
One problem you have is that you are specifying the mse and mingrad goals without considering the scale of the target. I find that reasonable goals are
MSEgoal = 0.01*mean(var(ttrn',1))
MinGrad = MSEgoal/ 200
0 Kommentare
Siehe auch
Kategorien
Mehr zu Deep Learning Toolbox finden Sie in Help Center und File Exchange
Produkte
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!