Skip to content

Latest commit

 

History

History
40 lines (27 loc) · 1.29 KB

README.rst

File metadata and controls

40 lines (27 loc) · 1.29 KB
Raw

Kian

Kian is the neural network designed to serve Wikidata.

This library in current shape adds statements to items based on categories in Wikis (Wikipedia, Wikisource, etc.) In order to add statements or parse articles you need to have pywikibot-core installed.

How to run this code for a run on classifying humans based on Persian Wikipedia:

python scripts/initiate_model.py -n faHuman -w fawiki -p P31 -v Q5
python scripts/train_model.py -n faHuman
python scripts/evaluate.py -n faHuman #To see AUC and fitness parameters
python scripts/parser.py -lang:fa -newpages:100 -n faHuman

Bottlenecks of speed in Kian are:

  1. Loading category links from Wikipedia. Since it caches them training different models from one wiki tends to work better
  2. Training the model. Since it's an ANN and ANNs are resource consuming to train, this may take a while but depends on Wiki you are working with.

Dependency injection

Kian is usable for any kind of training, you can simply inject the training set and get the result:

>>> from kian import Kian
>>> bot = Kian(training_set=the_training_set)
>>> bot.train()
>>> bot.theta
>>> bot.finalize(path='path_to_save_results')

Authors

Amir Sarabadani:
  • Ladsgroup<AT>gmail.com