Classifier evaluation (feature definition, feature size curve)
E = CLEVALFS(A,FEATDEF,CLASSF,FEATSIZES,LEARNSIZE,NREPS,S,TESTFUN)
| A|| Training dataset.|
| FEATDEF|| Untrained mapping(s), defining the feature space (possibly cell array)|
| CLASSF|| Untrained classifier(s) to be tested (possibly cell array)|
| FEATSIZES|| Vector of feature sizes (default: all sizes)|
| LEARNSIZE|| Number of objects/fraction of training set size (see GENDAT) |
| NREPS|| Number of repetitions (default: 1) |
| S|| Independent test dataset (optional)|
| TESTFUN|| Mapping,evaluation function (default classification error)|
| E|| Cell array [#FEATDEF,#CLASSF] with results (structures) See PLOTE for a description|
This routine is an extension of CLEVALF which makes a feature curve of a given dataset. CLEVALFS produces feature curves for every feature representation defined in one of the cells of FEATDEF (e.g. based on PCAM or FEATSELI). The result is a set of feature curves that can be plotted by PLOTE.
The following steps are taken
- Split data dataset A in a trainset and and testset S according to LEARNSIZE.
- Compute for an element in FEATDEF a feature representation. This includes a ranking of the features.
- Compute for this set of features and one of the classifiers in CLASSF a feature curve for the feature sizes as defined by FEATSIZES.
- repeat the last line for all classifiers in CLASSF.
- repeat the last three lines for all feature definitions in FEATDEF.
- Store all results as cells in E.
This function uses the RAND random generator and thereby reproduces only if its seed is saved and reset.
mappings, datasets, cleval, clevalf, testc, plote, gendat,
|This file has been automatically generated. If badly readable, use the help-command in Matlab.|