Cost-sensitive multiclass classification with Adaptive Regularization of Weights

This is a simple and efficient implementation of the Adaptive Regularization of Weights (AROW) algorithm for classification by Crammer et al. It is in python and it relies on the very efficient sparse vector implementation by Liang Huang which must be downloaded and compiled separately. While it is not as efficient as other implementations such as arowpp, it offers the following:

• multiclass classification
• cost-sensitive classification
• probability estimates

Download using this link or get it from github. In order to run it and see how the API is used, please download a version of the 20 newsgroups dataset available from LIBSVM and run (remember to install Liang Huang's sparse vector package first):

For any questions or bugs please contact me (andreas.vlachos AT cl.cam.ac.uk). If you find this software useful please acknowledge it. Thanks!