NEWS

---

kknn 1.3.2

OTHER CHANGES

• added native routine registration
• added some unit tests

BUG FIXES

• checks that k is smaller than the number of observations in the learning set
• if k was equal or larger the number of observation in the learning set result were spurious

kknn 1.3.1 (2016-03-26)

OTHER CHANGES

• predict got an additional argument "type"
• changed igraph code to version >=1.0
• train.kknn has a new argument ks, a vector specifying values of k

kknn 1.3.0 (2015-07-09)

OTHER CHANGES

• kknn is now hosted on github
• added travis-ci support

kknn 1.2-5 (2014-02-11)

BUG FIXES

• a browser() sneaked into kknn

kknn 1.2-4 (2014-02-11)

NEW FEATURES

• new function kknn.dist

OTHER CHANGES

• kknn returns now an index matrix of nearest neighbors

kknn 1.2-3 (2013-12-16)

BUG FIXES

• Namespace changed

kknn 1.2-2 (2013-03-13)

NEW FEATURES

• kknn has new parameter scale to switch off scaling
• new function cv.kknn for k-fold cross validation, replacing the function simulation in future
• default kernel changed to optimal

BUG FIXES

• interaction terms in formulas should be handled correctly

OTHER CHANGES

• speeded up kknn code a bit and reduced some memory allocations

kknn 1.2-1 (2012-10-30)

BUG FIXES

• solved some namespace problems

kknn 1.2-0 (2012-10-05)

NEW FEATURES

• added new kernel "optimal" to kknn Samworth (2012).

kknn 1.1-2 (2012-06-22)

BUG FIXES

• changed package to work with igraph >= 0.6

kknn 1.1-1 (2011-10-12)

OTHER CHANGES

• computation of Eucledian distances is 3-5 times faster.

kknn 1.1-0

NEW FEATURES

• added spectral clustering function specClust.

OTHER CHANGES

• computation of knn distances is much faster.

kknn 1.0-8 (2010-08-05)

BUG FIXES

• the data.frame for the training data set does not have to contain the response variable (suggested by Carlos J. Gil Bellosta).

kknn 1.0-5 (2007-10-11)

BUG FIXES

• A bug was fixed, leave-one-out crossvalidation now works fine.

kknn 1.0-3

NEW FEATURES

• new parameter ykernel: Window width of an y-kernel, especially for prediction of ordinal classes.

BUG FIXES

• A bug was fixed in the kernel="rank" in kknn.
• kernel = "gaussian" did not work correctly: this has been fixed.