1 个投票

Hello every one I am try to make neural network to predict permeability in oil field from wireline log so I have 5 input and one target I normaliz data to -1 and 1 but when I train network did not give me good r2 just 0.58 some time 0.62 what I do to reach r2 0.92 Many thanks muhammed

2 个评论
显示 无 隐藏 无

Greg Heath
Greg Heath 2014-10-28
Insufficient info.
size(input) [5 N ]
size(output) [ 1 N ]
Type of net fitnet ??
range of hidden Hmin:dH:Hmax ??
number of random weight trials for each value of H Ntrials ??
Any other non-default parameters??
muhammed kareem
muhammed kareem 2014-11-6
thanks dear Greg my input is 5*251 my target is 1*251......... i use newff ...... i do not know how range my hidden and weight>>>>>>>>>>>>>>
this is my code
net = newff(p, t, 20, {'logsig', 'purelin'});
net.divideParam.trainRatio = 75/100;
net.divideParam.testRatio = 15/100;
net.divideParam.valRatio = 10/100;
net.trainParam.epochs = 400;
net.trainParam.goal = 0.000001;
net.trainParam.max_fail = 200;
net.trainParam.lr = 0.06;
[net tr] = train(net,p,t);
a=sim(net,test);
postreg(a,tt);

请先登录，再进行评论。

请先登录，再回答此问题。

 采纳的回答

Greg Heath
Greg Heath 2014-11-7

1 个投票

The optimal value for H is usually obtained by trial and error. For each candidate value for H design Ntrials nets with different random initial weights.
I have posted many, many examples using a double for loop over H and weight initializations. Search NEWSGROUP and ANSWERS for the latest examples using
greg fitnet Ntrials
If you have the current functions FITNET (regression/curve-fitting)and PATTERNNET(classification/pattern-recognition) that automatically call FEEDFORWARDNET, use them instead of the OBSOLETE functions NEWFIT and NEWPR that automatically call NEWFF.
Hope this helps.
Thank you for formally accepting my answer
Greg

更多回答（0 个）

类别

在 帮助中心 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

产品

标签

Community Treasure Hunt

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

Start Hunting!

Translated by