Introduction

Per-- is a toolkit of Averaged Perceptron (Freund and Schapire, 1999) for path Labeling under the KISS principle (Keep It Simple and Stupid).

Features

• Simple and efficient
• Multiple usage: path labeling, path finding, sequence labeling and classification
• Can provide n-best outputs
• Can provide "marginal score"
• Can provide alpha (forward) values and beta (backward) values using forward-backward algorithm
• Open source

Download

Go there to download the binary files.

• per~~: executable file for path labeling
• per--: executable file for sequence labeling

Download the source code to build your own binary files on Windows or Linux.

Requirements

• gcc compiler (if you want to rebuild the toolkit)
• python3 (optional)

Usage

This data flow is the same with any other machine learning framework:

Data format

Learning and Predicting

An example for Chinese word segmentation

References

• Freund, Y. & Schapire, R. (1999). Large Margin Classification using the Perceptron Algorithm. In Machine Learning, 37(3):277-296.