UCI Optdigits数据的转化

文章目录

• 前言
• 一、原始数据情况
• 二、使用步骤
  • 1.引入数据
  • 2.按类标重组数据
• 总结

---

前言

UCI Dataset有很多原始数据，需要下载后修改格式为.mat。本文介绍optdigits的转换方法

---

提示：以下是本篇文章正文内容，下面案例可供参考

一、原始数据情况

UCI Machine Learning Repository: Optical Recognition of Handwritten Digits Data Sethttps://archive.ics.uci.edu/ml/datasets/Optical+Recognition+of+Handwritten+Digits

原始数据有四个文件     

   optdigits-orig.tra    Training        1934
    optdigits-orig.cv    Validation         946
    optdigits-orig.wdep    Writer-dependent     943
    optdigits-orig.windep    Writer-independent    1797

 

前三个合并为训练集，个数为3823.

二、使用步骤

1.引入数据

代码如下：

%optdigits-orig.tra	Training		1934
fileID = fopen('optdigits-orig.wdep','r');
formatSpec = '%s';
data = textscan(fileID,formatSpec);
data{1}(1:66)=[];
data = char(data{1});
% Define important parameters
nx = 32; ny = 32; nxny = nx*ny;
nNumbers = size(data,1)/ (nx + 1);
% Initialize the output
trainingMatrix = zeros(nxny, nNumbers);
output = zeros(nNumbers, 1);
% Subdevide data into images
for i = 1:nNumbers
    startingLocation = ((i-1)*(nx+1))+1;
    endingLocation = startingLocation + ny - 1; 
    outputLocation = endingLocation +1;
    trainingMatrix(:,i) = str2num(reshape(data(startingLocation:endingLocation, :), nxny, 1)); 
    output(i) = str2double(data(outputLocation, :));
end
% Visualize some the images
figure('Color', 'White');
for i = 1:6
    subplot(2,3,i)
    imagesc(reshape(trainingMatrix(:,i),32,32));
    axis equal; axis tight;
    set(gca,'xTickLabel', [], 'yTickLabel', []);
    colormap gray; 
end
train1=trainingMatrix;
train1cls=output;

%optdigits-orig.cv	Validation		 946
fileID = fopen('optdigits-orig.tra','r');
formatSpec = '%s';
data = textscan(fileID,formatSpec);
data{1}(1:65)=[];
data = char(data{1});
% Define important parameters
nx = 32; ny = 32; nxny = nx*ny;
nNumbers = size(data,1)/ (nx + 1);
% Initialize the output
trainingMatrix = zeros(nxny, nNumbers);
output = zeros(nNumbers, 1);
% Subdevide data into images
for i = 1:nNumbers
    startingLocation = ((i-1)*(nx+1))+1;
    endingLocation = startingLocation + ny - 1; 
    outputLocation = endingLocation +1;
    trainingMatrix(:,i) = str2num(reshape(data (startingLocation:endingLocation, :), nxny, 1)); 
    output(i) = str2double(data(outputLocation, :));
end
train2=trainingMatrix;
train2cls=output;

%optdigits-orig.wdep	Writer-dependent	 943
fileID = fopen('optdigits-orig.cv','r');
formatSpec = '%s';
data = textscan(fileID,formatSpec);
data{1}(1:65)=[];
data = char(data{1});
% Define important parameters
nx = 32; ny = 32; nxny = nx*ny;
nNumbers = size(data,1)/ (nx + 1);
% Initialize the output
trainingMatrix = zeros(nxny, nNumbers);
output = zeros(nNumbers, 1);
% Subdevide data into images
for i = 1:nNumbers
    startingLocation = ((i-1)*(nx+1))+1;
    endingLocation = startingLocation + ny - 1; 
    outputLocation = endingLocation +1;
    trainingMatrix(:,i) = str2num(reshape(data (startingLocation:endingLocation, :), nxny, 1)); 
    output(i) = str2double(data(outputLocation, :));
end
train3=trainingMatrix;
train3cls=output;
%generate traindata 1934+946+943=3823
tradata=[train1';train2']';
tradata=[tradata';train3']';
tradata=tradata';
letclasstra=[train1cls;train2cls];
letclasstra=[letclasstra;train3cls];

%optdigits-orig.windep	Writer-independent	1797
fileID = fopen('optdigits-orig.windep','r');
formatSpec = '%s';
data = textscan(fileID,formatSpec);
data{1}(1:66)=[];
data = char(data{1});
% Define important parameters
nx = 32; ny = 32; nxny = nx*ny;
nNumbers = size(data,1)/ (nx + 1);
% Initialize the output
trainingMatrix = zeros(nxny, nNumbers);
output = zeros(nNumbers, 1);
% Subdevide data into images
for i = 1:nNumbers
    startingLocation = ((i-1)*(nx+1))+1;
    endingLocation = startingLocation + ny - 1; 
    outputLocation = endingLocation +1;
    trainingMatrix(:,i) = str2num(reshape(data (startingLocation:endingLocation, :), nxny, 1)); 
    output(i) = str2double(data(outputLocation, :));
end
testdata=trainingMatrix;
testdata=testdata';
letclasstest=output;

2.按类标重组数据

代码如下

%rearrange data according to label 1..9
letclasstra=letclasstra+1;
letclasstest=letclasstest+1;
trcls=[]; tscls=[]; trdata=[]; tsdata=[]; ndigit=10;
for j=1:ndigit
    index=find(letclasstra==j);   trndata(j)=length(index);
    trcls=[trcls;letclasstra(index)]; 
    trdata=[trdata; tradata(index, :)];  tr{j}=tradata(index, :); 
   
    index=find(letclasstest==j); tsndata(j)=length(index);
    tscls=[tscls;letclasstest(index)];
    tsdata=[tsdata; testdata(index, :)];  ts{j}=testdata(index, :);   
end
save optdigits32a.mat trdata tsdata trcls tscls;
Xtr=reshape(trdata',[32 32 3823]);
Xts=reshape(tsdata',[32 32 1797]);
save optdigits32a_mx.mat Xtr Xts trcls tscls;

---

总结

这样，就可以将32位的数据改为标准的mat文件