OLaRankGreedy is an online solver of the dual formulation of support vector machines for structured output spaces. Here is displayed its implementation for sequence labelling using a greedy inference scheme. The per-sequence training time is the same as a perceptron based on the same inference procedure, up to a small multiplicative constant.
In comparison to existing algorithms, OLaRankGreedy matches the accuracy of batch solvers that use viterbi decoding after a single (much faster) pass over the training examples.
- Changes to previous version:
Initial Announcement on mloss.org.
No one has posted any comments yet. Perhaps you'd like to be the first?
Leave a comment
You must be logged in to post comments.