- Use file-backed blocking with dbm, dramatically increases size of data that can be handled without special programming
1.6.7
- Reduce memory footprint of matching
1.6.0
- Simplify .train method
1.5.5
- Levenshtein search based index predicates thanks to mattandahalfew
1.5.0
- simplified the sample API, this might be a breaking change for some - the active learner interface is now more modular to allow for a different learner - random sampling of pairs has been improved for linking case and dedupe case, h/t to MarkusShepherd
1.4.15
- frozendicts have finally been removed - first N char predicates return their entire length if length is less than N, instead of nothing - crossvalidation is skipped in active learning if using default rlr learner