Logistic Regression

The file data01.xml contains digits 0 and 1 of 20 samples each, collected on an Android device. Each of the collected images are of size 28 x 28 re-arranged to 1 x 784 matrix.

Sources:

Contents

Dataset

download data

dsFile = fullfile(mexopencv.root(), 'test', 'data01.xml');
if exist(dsFile, 'file') ~= 2
    disp('Download dataset...');
    url = 'https://cdn.rawgit.com/opencv/opencv/3.2.0/samples/data/data01.xml';
    urlwrite(url, dsFile);
end

load data

disp('Loading the dataset...');
S = cv.FileStorage(dsFile);
data = single(S.datamat);
labels = int32(S.labelsmat);
fprintf('%d rows of data\n', size(data,1));

Loading the dataset...
40 rows of data

split intro train/test sets

data_train = data(1:2:end,:);
data_test = data(2:2:end,:);
labels_train = labels(1:2:end);
labels_test = labels(2:2:end);
fprintf('training/testing samples count: %d / %d\n', ...
    size(data_train,1), size(data_test,1));

training/testing samples count: 20 / 20

display sample image

if mexopencv.require('images')  % MONTAGE
    showImage = @(data) montage(reshape(data', 28, 28, 1, []));
    subplot(121), showImage(data_train), title('train')
    subplot(122), showImage(data_test), title('test')
end

Classification

simple case with batch gradient

lr = cv.LogisticRegression();
lr.LearningRate = 0.001;
lr.Iterations = 10;
lr.Regularization = 'L2';
lr.TrainMethod = 'Batch';
lr.MiniBatchSize = 1;

fprintf('training... ');
tic
lr.train(data_train, labels_train);
toc

fprintf('predicting... ');
tic
responses = lr.predict(data_test);
toc

training... Elapsed time is 0.048445 seconds.
predicting... Elapsed time is 0.004559 seconds.

Prediction

show prediction report

disp('original vs predicted:');
disp([labels_test(:) responses(:)]);

if ~mexopencv.isOctave() && mexopencv.require('stats')
    %HACK: CONFUSIONMAT not implemented in Octave
    cm = confusionmat(labels_test, responses);
    disp('confusion matrix');
    display(cm);
end

acc = 100 * nnz(labels_test == responses) / size(responses,1);
fprintf('accuracy: %.2f%%\n', acc);

original vs predicted:
   0   0
   0   0
   0   0
   0   0
   0   0
   0   0
   0   0
   0   0
   0   0
   0   0
   1   1
   1   1
   1   1
   1   1
   1   1
   1   1
   1   1
   1   0
   1   1
   1   1
confusion matrix
cm =
    10     0
     1     9
accuracy: 95.00%

save the classfier

fname = fullfile(tempdir(),'NewLR_Trained.xml');
lr.save(fname);
disp(['classifier saved to: ' fname]);

classifier saved to: C:\Users\Amro\AppData\Local\Temp\NewLR_Trained.xml