Interactive demo of Support Vector Machines (SVM)

Note: you may have to disable your adblocker for this demo to work.

Toggle Clear Points

$\nu =$

Kernel: Linear Quadratic Radial Basis Functions Sigmoid $K(\mathbf x,\mathbf y) = \mathbf x\cdot\mathbf y$ $K(\mathbf x,\mathbf y) = \left(\gamma\mathbf x\cdot\mathbf y+c_0\right)^2$ $K(\mathbf x,\mathbf y) = \exp\left(-\gamma\|\mathbf x - \mathbf y\|^2\right)$ $K(\mathbf x,\mathbf y) = \tanh\left(\gamma\mathbf x\cdot\mathbf y+c_0\right)$

$\gamma =$     $c_0 =$

Highlight support vectors

Exception

Upps, something went wrong...

Hint

Add points to each of the two classes by left- and right-clicking inside the canvas to trigger the computation of a SVM model.

Usage

• Left- and right-click to add points belonging to either one of two classes to the canvas. Use the “Toggle” button to swap the classes associated with primary and secondary click if right-click is not an option (e.g. on mobile).
• At least one point in each class is necessary to “learn” an SVM model.
• Adjust the regularization parameter \(\nu\) between 0 and 1. Small values correspond to a hard margin, i.e. low tolerance towards misclassified data and may lead to overfitting of noisy data. Large values correspond to a soft margin and may ignore features in the data.
• Large values of \(\nu\) that are approaching 1 may cause the SVM optimization to fail.
• Select a kernel from the drop down menu. The definition of the currently selected kernel is displayed next to the drop down.
• As far as applicable, alter the parameters \(\gamma\) and \(c_0\) that enter the definition of the kernel.
• For the quadratic kernel, the decision boundary is a conic section.
• A modern browser is required, capable of executing WebAssembly.