Closed ajdapretnar closed 6 years ago
This dataset has continuous features and a discrete class. The only "out of the box" feature scoring method that is available is ANOVA. Other methods have to discretize the features behind the scene.
There's a problem with feature "ABHD11" - it is always nan (in fact, there's 860 such features). This could potentially trip up imputation, discretization or the feature scoring method.
We could adapt RemoveConstant for this purpose.
Orange version
3.15.dev
Expected behavior
Rank works reasonably fast on computers with sufficient RAM.
Actual behavior
Rank is slow on some machines. It worked fine on a PC with 2GB RAM, while it took forever on two with 8GB RAM. All machines are Windows, but it is probably not OS dependent.
Steps to reproduce the behavior
Try to rank blood-loneliness.tab with 19335 features.
Additional info (worksheets, data, screenshots, ...)