DevRetreat23ProjectQuantitativeTesting

Jump to bottom

Andrew Howe edited this page Nov 6, 2023 · 3 revisions

Implementing a testsuite to test rules for FPs on a scale

Description / Goals

Quantitative approach to false positives by testing rules (transformations + operator (regex, parallel match, etc.)) against large bodies of examples of natural language
Bodies of text could be Wikipedia, Reddit, news articles (example corpora: https://wortschatz.uni-leipzig.de/en/download)
Depending on results, this could be used to optimize CRSv4 before the release

Participants

@M4tteoP
@theseion
@redxanadu
@airween
@franbuehler

Ideas

What target(s) do we want to examine?
- ARGS, because this covers most relevant rules and also is where we see natural language occurring?