- Fix a lot of bugs present in the older Python version, - add the option to load weights in the sparse, map, or dense format, - add Windows compatibility.
0.5.2
- Add more datasets from XML repo, - fix some bugs in OVR and BR training.
0.5.1
- Fix `ndcg_at_k` method, - add new propensity scored measures: `psrecall_at_k`, `psdcg_at_k`, `psndcg_at_k`, - improve performance of `load_libsvm` method.
0.5.0
- Add prediction with weights, - add ofo method to Python bindings, - measures at k now return an array of all values from 1 to k, - fix many minor bugs.