2011/10/26 SK Sn <[email protected]>: > Hi there, I am looking into how to implement multi-label > text classification. > Generally, am training models just as normal multi-class classification, but > for predication/test, I want to make it multi-label, i.e. getting more than > one labels for predication on a given test item. > > The easiest way would be implementing a battery of one-vs-all classifiers, > which actually as I understand, is what happens under the hood when one uses > most of the classification methods for multi-class.
Yes. > However, what I can not figure out is that how can I manipulate the process > of prediction in order to generate multi-label rather than only one label as > the prediction result. > Are there any provided APIs to do the multi-label classification, or I need > to build the multi-label classification from the ground by building the set > of classifiers by myself? There is (yet) no official scikit-learn API for building multi-label classifiers so I would advise you to come up with your own wrapper utilities. The following LabelBinarizer class might help you turn one representation to another though: https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/preprocessing/__init__.py#L437 > A final off-topic question is, as I noticed this mailing list is more for > developing coordination, am I suppose to post question like this to > stackoverflow or some other places? It's ok to use the mailing list for both developer discussions and user questions. Some developers also lurk around stackoverflow, cross validated (http://stats.stackexchange.com/ more focused on theoretical stats and parametrics models) and http://metaoptimized.com/qa (more focused on machine learning, graphical models, deep learning & neural nets and practical applications such as NLP and computed vision). > Also, communications in mailing list > seems to be not SEO optimized so that others hardly can search and get > to the contents in the mailing list. This may be a waste of resources. Indeed the sourceforge web interface for the mailing list archives is very bad. Maybe we should think about switching to something else? Maybe librelists.com or google groups? The web access for librelist archives is pretty nice, e.g. for the joblib project: http://librelist.com/browser/joblib/ -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel ------------------------------------------------------------------------------ The demand for IT networking professionals continues to grow, and the demand for specialized networking skills is growing even more rapidly. Take a complimentary Learning@Cisco Self-Assessment and learn about Cisco certifications, training, and career opportunities. http://p.sf.net/sfu/cisco-dev2dev _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
