- Block indexes can now be persisted by using the index=True argument in the writeSettings method
1.4.1
- Now uses C version of double metaphone for speed - Much faster compounding of blocks in block learning
1.4.0
- Block learning now tries to minimize the total number of comparisons not just the comparisons of distinct records. This decouples makes block learning from learning classifier learning. This change has requires new, different arguments to the train method.
1.3.8
- Console labeler now shows fields in the order they are defined in the data model. The labeler also reports number of labeled examples - `pud` argument added to the `train` method. Proportion of uncovered dupes. This deprecates `uncovered_dupes` argument
1.3.0
- If we have enough training data, consider Compound predicates of length 3 in addition to predicates of length 2
1.1.1
- None now treated as missing data indicator. Warnings for deprecations of older types of missing data indicators