Package now depends on R > 3.5.0, new imports of pillar, utils
fix warning when misvm() has matrix passed
fix .reorder() ambiguity
pass lintr checks
re-work internals for easier testing
v0.3.1
Fix bug where NaN columns passed to mildsvm() would fail
Fix bug where columns with identical values passed to mildsvm() would fail
v0.3.0
Add new method to mildsvm(): method = ‘qp-heuristic’. This works similar to the method of the same name in misvm(), but uses the SMM kernel from kme() in the underlying calculations.
Fix bug in classify_bags() when using factors
v0.2.0
The main modeling functions (misvm(), mildsvm(), and smm()) now have three methods:
Formula method (i.e. misvm(mi(y, bags) ~ x1 + x2, data = df, …))
Data-frame method (i.e. misvm(x, y, bags, …))
Method for the mild_df class (I.e. misvm(mil_data, …)). This method often performs non-trivial aggregation or transformation since misvm() and smm() work naturally on MIL data and supervised data, respectively.
Prediction on main modeling functions always returns a tibble with a single column depending on the type argument
Kernel feature maps functions are now organized as kfm_nystrom(), kfm_exact() with a build_fm() method.
Update MilData class to mild_df class, and improve the class methods and constructors.
Many internal methods removed and restructured.
v0.1.0
Initial release. This release has several known bugs and an early input/output scheme that has since been revised. This represents a mostly working starting point.