Kazu as a Library

We understand that many use cases for NER and Linking tend to come as part of a larger data processing pipeline, and thus it makes sense to use Kazu as a library of code. This is generally easy to do, with the following caveats:

Dependency conflicts

Capping dependencies for a library’s requirements is a controversial aspect of development.

We try to keep kazu as flexible as possible with regard to dependencies. This means we don’t cap dependencies, and this can sometimes cause unforeseen errors. To this end, we test kazu with the latest version of each of its dependencies as descibed in the pyproject.toml file. If you suspect you are having dependency clash issues, you can view the dependencies a given Kazu model pack was tested with via the tested_dependencies.txt file (located at the top level of a model pack). Try installing the version of the problematic dependency listed here.