Using decision objects in the Pattern Recognition Toolbox
prtDecision objects are intended to be used in algorithms with, or as part of prtClass or prtAlgorithm objects. They change the output of the run, crossvalidate and kmeans functions from decision statistics to class labels. They also determine the operating point that the decision is made at.
Contents
prtDecision objects as the internalDecider of a prtClass object
The simplest way to use a prtDecision object is by setting the internalDecider property of a prtClass object. For example, to set the operating point so that the classifier has the minimum probabilty of error:
ds = prtDataGenBimodal; % Load a data set classifier = prtClassKnn; % Create a clasifier % Set the internal decider classifier.internalDecider = prtDecisionBinaryMinPe; result = classifier.kfolds(ds,10); % K-folds cross validate pc = prtScorePercentCorrect(result) % Check the percent correct
pc =
0.7900
Other valid binary decision objects are the prtDecisionBinarySpecifiedPf and prtDecisionBinarySpecifiedPd objects, which force the classifier to operate at a specific Pf or Pd.
prtDecision objects as part of a prtAlgorithm
prtDecisions can also be part of prtAlgorithms, the operation is very similar. For example, the following implements the same classifier as above:
alg = prtClassKnn + prtDecisionBinaryMinPe; % Create an algorithm object result = classifier.kfolds(ds,10); % K-folds cross validate pc = prtScorePercentCorrect(result) % Check the percent correct
pc =
0.7900
Note, the percent correct in the two examples may vary slightly due to the inherent randomness of kfolds cross validation
M-ary decisions
M-ary decisions can be performed using the prtDecisionMap object:
ds = prtDataGenMary; classifier = prtClassKnn; classifier.internalDecider = prtDecisionMap; result = classifier.kfolds(ds,10); % K-folds cross validate pc = prtScorePercentCorrect(result) % Check the percent correct
pc =
0.8367
All prtDecision objects in the Pattern Recognition Toolbox have the same API as discussed above. For a list of all the different functions, and links to their individual help entries, A list of commonly used functions