RSMExplain is doing 4 standard deviations from the mean clamping to remove outlier. This works fine for models trained by RSMTool, however it doesn't make sense for SKLL models. It would be nice to have a config setting to turn clamping on and off.