Package

weka.classifiers.functions

Synopsis

Implements the stochastic variant of the Pegasos (Primal Estimated sub-GrAdient SOlver for SVM) method of Shalev-Shwartz et al. (2007). This implementation globally replaces all missing values and transforms nominal attributes into binary ones. It also normalizes all attributes, so the coefficients in the output are based on the normalized data. Can either minimize the hinge loss (SVM) or log loss (logistic regression). This implementation can be trained incrementally on (potentially) infinite data streams. For more information, see

S. Shalev-Shwartz, Y. Singer, N. Srebro: Pegasos: Primal Estimated sub-GrAdient SOlver for SVM. In: 24th International Conference on MachineLearning, 807-814, 2007.

Available in Weka 3.6.x - 3.7.1. Available via the package management system for Weka >= 3.7.2 (SPegasos).

Options

The table below describes the options available for SPegasos.

Option	Description
debug	If set to true, classifier may output additional info to the console.
dontNormalize	Turn normalization off
dontReplaceMissing	Turn off global replacement of missing values
epochs	The number of epochs to perform (batch learning). The total number of iterations is epochs * num instances.
lambda	The regularization constant. (default = 0.0001)
lossFunction	The loss function to use. Hinge loss (SVM) or log loss (logistic regression).

Capabilities

The table below describes the capabilites of SPegasos.

Capability	Supported
Class	Binary class, Missing class values
Attributes	Binary attributes, Unary attributes, Numeric attributes, Missing values, Empty nominal attributes, Nominal attributes
Min # of instances	0