Matlab - neural networks - How to use different datasets for training, validation and testing? -

Matlab - neural networks - How to use different datasets for training, validation and testing? -

best

i've question neural networks in matlab.

first of all, i've little nn, 2 inputs, 1 hidden layer 10 neurons , 1 output. , works fine. question i've is. can determine training date, validation info , test data?

i know, if utilize e.g. net = feedforwardnet(10); can split overall dataset e.g.70/100 15/100 , 15/100. don't want this, because in case want train nn 1000 data-points, validate them data-points , utilize independent data-set of 1000 data-points test them. other words, want command these 3 interdependent data-sets.

thus, can help me?

kind regards

edit, don't want utilize data-set 3000 data-points , set devideparams on 1/3 1/3 & 1/3.

best myself

when utilize feedforwardnet can define split parameters

net.divideparam.trainratio = 1/3; net.divideparam.valratio = 1/3; net.divideparam.testratio = 1/3;

you know info divided 3 pieces. (i) didn't know data.

but when , i, train network via next command line:

[net,tr]=train(net,x,t);

then, tr contain necessary info, illustration :

tr.trainind 1x1000 double, tr.valind 1x1000 double, tr.testind 1x1000 double,

thus, e.g. tr.trainind, contain indexes of our set of info used training. also, in tr, can see type of tr.dividefcn set on dividerand means indexes picked @ random. logical, there possibility indexes aren't picked randomly means that, if combine both things. should possible utilize test set --> net.divideparam.testratio = 0 , utilize 2 different train , validation sets --> net.divideparam.trainratio = 1/2 , net.divideparam.valratio = 1/2 - if can set tr.dividefcn on chronological, sequential. lastly not least, if possible have nil more do, set training , validation set 1 info set, etc...

kind regards myself

matlab neural-network