- Remove training data from coverage benchmarks
- Make propositional benchmark use crossvalidation
- Add coverage benchmarks to relational and propositional
- Add syllogistic network models (2019-Riesterer)
- Improve package documentation
- Overhaul benchmark handling
- Add possibility to supply multiple datasets as list in benchmark JSON
- Streamline evaluator
- Add "prediction" evaluation type as an evaluation without adaption
- Change model interface to pre_train, pre_train_person, and pre_train_background
- Allow "%ccobra%" placeholder to occur in all benchmark paths
- CCOBRA now stores the result html on disk and opens this file in the browser instead of temporarily creating a HTTP server itself
- Add generalized syllogistic helper functionality