knnB {MLInterfaces} | R Documentation |
This document describes a family of wrappers of calls to machine learning classifiers distributed through various R packages. This particular document concerns the classifiers for which training-vs-test set application makes sense.
For example, knnB
is a wrapper for a call to knn
for objects
of class ExpressionSet
. These interfaces, of the form [f]B
provide a common calling
sequence and common return value for machine learning code in function [f]
.
For details on the additional arguments that may be passed to any covered
machine learning function f
, check the manual page for that function.
This will require loading the package in which f
is found.
knnB(exprObj, classifLab, trainInd, k = 1, l = 1, prob = TRUE, use.all = TRUE, metric = "euclidean") # # for such functions as nnetB, use the same first three # parameters, and then add optional parameters from the nnet API #
exprObj |
An instance of the exprset class. |
classifLab |
The name of the phenotype variable to use for classification. |
trainInd |
integer vector: Which elements are the training set. |
k |
The number of nearest neighbors. |
l |
See knn for a complete description. |
prob |
See knn for a complete description. |
use.all |
See knn for a complete description. |
metric |
See knn for a complete description. |
See knn
for a complete description of
parameters to and details of the k-nearest neighbor procedure
in the class
package.
For other interfaces, such as ldaB, nnetB, rpartB, gbmB, randomForestB, and so on, see the usage note above and also see the man pages for those functions. For each of these functions you will need to attach the appropriate package in order to examine the man page.
The MLearn
interface is a more unified approach but is still maturing.
An object of class "classifOutput"
This class unifies the representation of results of machine
learning algorithms implemented by different designers.
Jess Mar, VJ Carey <stvjc@channing.harvard.edu>
xval
for information on how
to obtain various cross-validated fits, and MLearn
for a less fragmented implementation of the interface.
# access and trim an ExpressionSet library(golubEsets) data(Golub_Merge) smallG <- Golub_Merge[1:60,] # set a PRNG seed for reproducibilitiy set.seed(1234) # needed for nnet initialization # now run the classifiers knnB( smallG, "ALL.AML", 1:40 ) nnetB( smallG, "ALL.AML", 1:40, size=5, decay=.01 ) lvq1B( smallG, "ALL.AML", 1:40 ) naiveBayesB( smallG, "ALL.AML", 1:40 ) svmB( smallG, "ALL.AML", 1:40 ) baggingB( smallG, "ALL.AML", 1:40 ) ipredknnB( smallG, "ALL.AML", 1:40 ) sldaB( smallG, "ALL.AML", 1:40 ) ldaB( smallG, "ALL.AML", 1:40 ) qdaB( smallG[1:10,], "ALL.AML", 1:40 ) pamrB( smallG, "ALL.AML", 1:40 ) rpartB( smallG, "ALL.AML", 1:35 ) randomForestB( smallG, "ALL.AML", 1:35 ) gbmB( smallG, "ALL.AML", 1:40, n.minobsinnode=3 , n.trees=6000) stat.diag.daB( smallG, "ALL.AML", 1:40 )