Is it possible to get the equation of the separating line generated when using SVMTRAIN to train a classifier using a linear kernel function in Bioinformatics Toolbox 3.5 (R2010a) for a 2D set of data?

I am using the SVMTRAIN function to train a classifier for a 2D set of data using a linear kernel function (the default). When the 'showplot' option is set to 'true', a plot is displayed showing the grouped data and separating line for the classifier. I would like a way to obtain the equation for this line in the form y=mx+b.

Answer 1

[MathWorks Support Team]

The ability to obtain the equation of the classifying line drawn by SVMTRAIN is not available in Bioinformatics Toolbox.

To work around this issue for a 2D set of data, you can obtain the data points of the line object after it has been plotted, and calculate the y-intercept and slope of the line using these points.

% Generate training data (x1,y1 are positive classification, x2,y2 are

% negative classification)

x1 = randn(100,1);

y1 = randn(100,1);

x2 = randn(100,1) + 3;

y2 = randn(100,1) + 3;

trueDecision = [zeros(size(x1))  ;  ones(size(x2))];

Training = [x1 y1 ; x2 y2];

Group = trueDecision;

% Train the classifier

svmOutput = svmtrain(Training, Group, 'ShowPlot', true);

% Get the handle to the classifying line

axesHandle = svmOutput.FigureHandles{1};    % Get the handle to the axes

childHandles=get(axesHandle,'children');    % Find the axes' children

hggroupHandle=childHandles(1);              % Get HGGROUP object handle that contains the line

lineHandle=get(hggroupHandle, 'children');  % Get the line's handle

% Get the X and Y data points of the line

xvals=get(lineHandle, 'XData');

yvals=get(lineHandle, 'YData');

% Using 2 points on the line, calculate slope and y-intercept

m=(yvals(2)-yvals(1))/(xvals(2)-xvals(1))

b=yvals(1)-m*xvals(1)

% Plot the y=mx+b line to check it (blue dashed line)

hold on;

plot(xvals, m*xvals+b, '--', 'LineWidth', 3);