Closed GoogleCodeExporter closed 9 years ago
This is wrong. train_frac must be set to 0 in order for TrainingSet->split()
to make balanced training sets. If its not 0, then it will use the ratio
per-class, and ignore the testing/training images setting!
I realize that the debug line is completely deceptive as to what will happen.
check_split_params() sets train_frac_p to 0 even if you specified a -r (without
the '#' decorator) because the default is balanced, which means split() uses
the straining images and testing images instead of the fraction. If the user
specified a -r (instead of -i and -j), check_split_params() simply sets the -i
and -j params for you using the -r you gave. So the fraction is always 0
unless you have -r#, which is the intent.
Original comment by i...@cathilya.org
on 18 Feb 2011 at 2:49
OK, so how do I mark this as "fixed" or "not a bug"?
Original comment by i...@cathilya.org
on 18 Feb 2011 at 2:50
I fixed the output to only report the frac when its actually in play (i.e. when
-r# is set).
Original comment by i...@cathilya.org
on 19 Feb 2011 at 4:45
Original issue reported on code.google.com by
christop...@nih.gov
on 18 Feb 2011 at 1:16