This release brings much more robustness ensured through extensive JUnit testing. Furthermore, we improved the usability for creating ML-Plan instances and configuring it for various problems.
As of this release ML-Plan can be used for the following combinations of learning problems and backends:
- Standard classification - WEKA
- Standard classification - scikit-learn
- Regression - WEKA
- Regression - scikit-learn
- Multi-label classification - MEKA
- Remaining Useful Lifetime Estimation - tsfresh & scikit-learn
For some of these combinations we have also introduced alternative configuration setups to work with different search spaces etc. Another novelty concerns the use of different internal validation performance measures. These can be configured now reliably without causing issues for the search algorithm nor the selection phase.
Working with scikit-learn via the command line is now much more flexible due to the option of using environments. Moreover, the speed of the ScikitLearnWrapper has been improved significantly by tweaking the dataset reading procedure a lot.