SPegasos
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 |