In MATLAB while using Neural fitting toolbox (nftool) for solving input-output mapping problem we have to divide samples for “Training, Validation and Testing”. For developing the network we need to train the samples. But if we retrain the samples nftool takes new samples every time. Is there any way by which we can fix the samples for “Training, Validation and Testing”? I don’t want to change my samples every time I run my nftool code.
Dear Mr. Borate,
In MATLAB, the default selection is "dividerand" means randomly divided data sets. Instead of "dividerand", you may use "divideind" selection that you can divide your data by their index.
you may see detailed information in MATLAB "help divideind" command.
net.divideParam.trainInd
net.divideParam.valInd
net.divideParam.testInd
Dear Mr. Borate,
In MATLAB, the default selection is "dividerand" means randomly divided data sets. Instead of "dividerand", you may use "divideind" selection that you can divide your data by their index.
you may see detailed information in MATLAB "help divideind" command.
net.divideParam.trainInd
net.divideParam.valInd
net.divideParam.testInd
choose the divideind, to divide you data instead the default selection dividerand
I would suggest the command line scripts , you can control each net and its characteristics , even the choice of algorithms.
I think you can use GUI tool Neural Network Toolbox MATLAB by using the command line interface and use following:
net.divideParam.trainInd
net.divideParam.valInd
net.divideParam.testInd
net.divideParam.trainInd
net.divideParam.valInd
net.divideParam.testInd
- Badges
- Science topic