- webstruct.model.NER now uses
requestslibrary to make HTTP requests;
- changed default headers used by webstruct.model.NER;
webstruct.infer_domainmodule useful for proper cross-validation;
- webstruct.webannotator.to_webannotator got an option to add
<base>tag with the original URL to the page;
- fixed a warning in webstruct.gazetteers.geonames.read_geonames;
- add a few more country names to countries.txt list.
- fixed a bug in NER.extract().
- sklearn-crfsuite is used as a CRFsuite wrapper, CRFsuiteCRF class is removed;
- comments are preserved in HTML trees because recent Firefox puts
<base>tags to a comment when saving pages, and this affects WebAnnotator;
- fixed ‘dont_penalize’ argument of webstruct.NER.extract_groups_from_url;
- new webstruct.model.extract_entity_groups utility function;
- HtmlTokenizer and HtmlToken are moved to their own module (webstruct.html_tokenizer);
- test improvements;
There are many changes from previous version: API is changed, Python 3 is supported, better gazetteers support, CRFsuite support, etc.