How could I cancel pre-processing and post-processing stages when training a neural network?

Whenever I use fitnet it applies minmax to my data by default, however my data is already scaled, so I was just wondering if there was a way to change this default feature.

Answer 1

[Greg Heath]

To remove mapminmax as a processFcn but still keep removeconstantrows

close all, clear all, clc

% WITH DEFAULT MAPMINMAX

 [x,t] = simplefit_dataset;
 net = fitnet;
 rng(0)
 [net tr y e ] = train(net,x,t);
 NMSE1 = mse(e)/mean(var(t',1))  % 1.7558e-05

% WITHOUT DEFAULT MAPMINMAX

 [x,t] = simplefit_dataset;
 net = fitnet;
 net.inputs{1}.processFcns={'removeconstantrows'};
 net.outputs{2}.processFcns={'removeconstantrows'};
 rng(0)
 [net tr y e ] = train(net,x,t);
 NMSE2 = mse(e)/mean(var(t',1))  % 1.7241e-05

Hope this helps.

Thank you for formally accepting my answer

Greg

[Alexandra Ntemourtsidou]

Thank you so much for you response, I was just wondering if this cancels post-processing as well. Also could I remove 'remove const rows'? So have no processing features whatsoever?

[Greg Heath]

Yes. The first statement is for inputs and the second for outputs.

If you do not want to eliminate constant rows, use '' for the RHS (maybe {} will work also)

Hope this helps.

Greg