Opened 16 years ago

Closed 16 years ago

#271 closed defect (fixed)

NBC trains incorrectly when have total zero weight for a feature

Reported by: Peter Owned by: Peter
Priority: minor Milestone: yat 0.4
Component: classifier Version: trunk
Keywords: Cc:


When there is no data for a feature m is set to 0 and sigma is set to zero. We should set this to NaN and take care of the NaN in predict.

Change History (1)

comment:1 Changed 16 years ago by Peter

Resolution: fixed
Status: newclosed

Solved in [960] and [961]. If the number of valid data points is 1 or smaller for a specific feature and class, then that feature is ignored for the prediction of that class.

Note: See TracTickets for help on using tickets.