Closed GoogleCodeExporter closed 9 years ago
Original comment by steven.b...@gmail.com
on 24 Jul 2012 at 5:34
Original comment by lee.becker
on 17 Feb 2013 at 5:57
This issue was closed by revision db071ff2028f.
Original comment by steven.b...@gmail.com
on 19 Feb 2013 at 1:23
The fix for this issue includes two potentially incompatible changes: (1) the
LIBLINEARBooleanOutcomeXXX classes now inherit from different superclasses and
have slightly different internal APIs, and (2) a new LIBLINEAR feature encoder
was introduced.
(1) should only affect code that inherited from LIBLINEARBooleanOutcomeXXX and
code that simply used the LIBLINEARBooleanOutcomeXXX public APIs should be
unaffected
(2) should produce the same String-to-int feature mappings as the previous
feature encoder, but it's probably still safer to retrain any existing
LIBLINEAR models just in case
Original comment by steven.b...@gmail.com
on 19 Feb 2013 at 1:30
I had to make an additional change in revision
d4137f3f6fe37a78bc57f58805c1ca4bbdd2231a that adds a 'bias' feature to the
LIBLINEAR feature encoder. This is necessary for LIBLINEAR to be able to
predict the majority class for instances consisting entirely of features not
seen during training.
So any existing LIBLINEAR models will definitely need to be retrained as part
of this change.
Original comment by steven.b...@gmail.com
on 19 Feb 2013 at 6:04
This issue was closed by revision db071ff2028f.
Original comment by steven.b...@gmail.com
on 28 Mar 2013 at 5:53
Original issue reported on code.google.com by
steven.b...@gmail.com
on 5 Apr 2012 at 9:20