/nlg-metrics

Primary LanguagePythonMIT LicenseMIT

Blanche library

made-with-python

Blanche is a library containing the most important nlg evaluation metric and some emerging ones

List of supported metrics

N-gram overlap

Model-based

QA based

All metrics have citations in blanche source code

Note: FEQA and FactCC may have some trouble in execution

EXAMPLE OF USE:

From blanche/src folder

image

UPDATE TEST SET

  • Create data folder (blanche/data)
  • Put it file of references and predictions update_test_set("name of test set", "references file name", "predictions file name")
  • Results are saved in a metrics directory name_of_test_set_name_metrics