Classes
class	CSVRLight
Defines
#define	IGNORE_IN_CLASSLIST
	Class SVRLight, performs support vector regression using SVMLight.

Define Documentation

#define IGNORE_IN_CLASSLIST

Class SVRLight, performs support vector regression using SVMLight.

The SVR solution can be expressed as

$f({\bf x})=\sum_{i=1}^{N} \alpha_i k({\bf x}, {\bf x_i})+b$

where $\alpha$ and $b$ are determined in training, i.e. using a pre-specified kernel, a given tube-epsilon for the epsilon insensitive loss, the follwoing quadratic problem is minimized (using the chunking decomposition technique)

$\begin{eqnarray*} \max_{{\bf \alpha},{\bf \alpha}^*} &-\frac{1}{2}\sum_{i,j=1}^N(\alpha_i-\alpha_i^*)(\alpha_j-\alpha_j^*){\bf x}_i^T {\bf x}_j -\sum_{i=1}^N(\alpha_i+\alpha_i^*)\epsilon - \sum_{i=1}^N(\alpha_i-\alpha_i^*)y_i\\ \mbox{wrt}:& {\bf \alpha},{\bf \alpha}^*\in{\bf R}^N\\ \mbox{s.t.}:& 0\leq \alpha_i,\alpha_i^*\leq C,\, \forall i=1\dots N\\ &\sum_{i=1}^N(\alpha_i-\alpha_i^*)y_i=0 \end{eqnarray*}$

Note that the SV regression problem is reduced to the standard SV classification problem by introducing artificial labels $-y_i$ which leads to the epsilon insensitive loss constraints *

$\begin{eqnarray*} {\bf w}^T{\bf x}_i+b-c_i-\xi_i\leq 0,&\, \forall i=1\dots N\\ -{\bf w}^T{\bf x}_i-b-c_i^*-\xi_i^*\leq 0,&\, \forall i=1\dots N \end{eqnarray*}$

with $c_i=y_i+ \epsilon$ and $c_i^*=-y_i+ \epsilon$

This implementation supports multiple kernel learning, i.e. if a CCombinedKernel is used the weights $\beta$ in $k_{combined}({\bf x}, {\bf x'}) = \sum_{m=0}^M \beta_m k_m({\bf x}, {\bf x'})$ can be determined in training (cf. Large Scale Multiple Kernel Learning Sonnenburg, Raetsch, Schaefer, Schoelkopf 2006).

linadd optimizations were implemented for kernels that support it (most string kernels and the linear kernel), which will result in significant speedups.

Definition at line 62 of file SVRLight.h.

SVRLight.h File Reference

Classes

Defines

Define Documentation