K-cross validation ANN

2 次查看（过去 30 天）
Mallikarjun Yelameli
评论： Greg Heath 2017-5-9
Check my following code, if there is any error (may be logical) let me know. Thank You.
clc;clear all; close all;
%training data of size (1024x3600) It has 1024 features and 3600 samples
x=dlmread('train_data.txt');
%targets data labels of size (15x3600) The data has 15 classes.
y=dlmread('targets_train.txt');
%test data has 1024 features and 900 samples.
test_data=dlmread('test_data.txt');
%target test data is of size (15x900);
test_target=dlmread('test_target.txt');
% Take note that, the train and test data is already divided. These two datasets are independant.
best_acc=0; %this is the best cross-validation accuracy, initialized as zero.
nr_fold=5; % number of cross-validation required, in my case, I took number 5
%number hidden layers iterates over the loop, to calculate the best best hidden layer.
hiddenLayerSize=10:1:20;
for i=1:length(hiddenLayerSize)
net=patternnet(hiddenLayerSize(i));
net.divideFcn='';
acc=get_cv_ac(y,x,nr_fold,net); %get_cv_ac function is written below.
if acc>best_acc
best_acc=acc; best_hiddenLayer=hiddenLayerSize(i);
end
fprintf('%g %g (best acc=%g, best hidden Layer=%g)\n)', acc,hiddenLayerSize(i), best_acc, best_hiddenLayer);
end
end
%now retrain the whole training data with best hiddenLayer number.
net=patternnet(best_hiddenLayer);
net.divideFcn='';
best_model=ovrtrain(net,x,y); %ovrtrain function is written below.
%now check the performance of test dataset (this independant) on this best_model
[pred,pred_ind,actual_ind,acc]=ovrpredict(best_model,test_data,test_target); %ovrpredict function is written below.
fprintf('Accuracy on test dataset=%g\n',acc)
Now get_cv_ac function code
function [ac]=get_cv_ac(y,x,nr_fold,net)
%y is train label
%x is train data
len=size(y,2);
ac=0;
rand_ind=randperm(len);
for i=1:nr_fold
val_ind=rand_ind([floor((i-1)*len/nr_fold)+1:floor(i*len/nr_fold)])';
train_ind=rand_ind[1:len]';
train_ind(test_ind)=[];
[model,tr]=ovrtrain(net,x(:,train_ind),y(:,train_ind)); %ovrtrain function is written below
[pred,pred_ind,actual_ind,acc]=ovrpredict(model,x(:,val_ind),y(:,val_ind)); %ovrpredict function is written below
ac=ac+sum(pred_ind==actual_ind);
end
ac=ac/len*100;
fprintf('cross-val accuracy=%g\n',ac);
end
Now ovrtrain function is written below
function [model,tr] = ovrtrain(net,x,y)
net=configure(net,x,y);
[model,tr]=train(net,x,y);
end
Now ovrpredict function is written below
function [pred,pred_ind,actual_ind,acc]=ovrpredict(model,x,test_ind)
pred=model(x);
[values,pred_ind]=max(pred,[],1);
[~,actual_ind]=max(test_ind,[],1);
acc=(sum(pred_ind==actual_ind)/size(x,2))*100
end
  1 个评论
Greg Heath
Greg Heath 2017-5-9
Run your code. If there are any errors, let us know.
Greg

请先登录，再进行评论。

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Sequence and Numeric Feature Data Workflows 的更多信息

标签

产品

Community Treasure Hunt

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

Start Hunting!

Translated by