|
background: verifying hundreds of millions of documents turned out to be
a bit slow; anecdata: running clustering and verification over 1.8B
inputs tooks over 50h; cf. the Go port (skate) required about 2-4h for
those operations. Also: with Go we do not need the extra GNU parallel
wrapping.
In any case, we aim for fuzzycat refactoring to provide:
* better, more configurable verification and small scale matching
* removal of batch clustering code (and improve refcat docs)
* a place for a bit more generic, similarity based utils
The most important piece in fuzzycat is a CSV file containing hand
picked test examples for verification - and the code that is able to
fulfill that test suite. We want to make this part more robust.
|