svm-scale manpage contains misleading information

Bug #278166 reported by Zeno Gantner on 2008-10-04
2
Affects Status Importance Assigned to Milestone
libsvm (Ubuntu)
Undecided
Unassigned

Bug Description

Binary package hint: libsvm-tools

The manpage to svm-scale contains the following lines which explain the -s and -r options:

      -s save_filename
              save_filename indicates the filename to save the scaled data to.

       -r restore_filename
              restore_filename indicates the filename reserved to hold the
              original (unscaled) data in case there is a need to restore.

In reality, the scaled data is written to STDOUT, and the range information for each feature is written to save_filename.
This range information can then be used in order to use the same scaling function for a different data set (typically the test data), which is done with the -r parameter.

For example:
svm-scale -l 0 -u 1 -s range training.data > training.data_scaled
svm-scale -l 0 -u 1 -r range test.data > test.data_scaled

To fix this issue, I suggest to add the sentence

"The scaled data is written to STDOUT." to the section DESCRIPTION
and replace the option descriptions above by

      -s save_filename
              save_filename indicates the filename to save the scaling parameters to.

       -r restore_filename
              restore_filename indicates the filename reserved to hold the
              scaling parameters in case there is a need to use this information
              to perform the same scaling operation on a similar dataset.

Best regards,
  Zeno

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers