| PRTools contents |
LOSO
[E,C,D] = LOSO(A,CLASSF,LABLISTNAME)
[E,C,D] = LOSO(A,CLASSF,SET_LABELS)
[E,C,D] = LOSO(A,CLASSF,SET_LABELS,SET_LABLIST)
| Input | |
| A | Dataset |
| CLASSF | Untrained classifier |
| LABLISTNAME | Name of label list in case of multiple labeling |
| SET_LABELS | Set of labels for objects in A |
| SET_LABLIST | Order and selection of labels in SET_LABELS |
| Output | |
| E | Classification error |
| C | Array with numbers of erroneaously clasified objects per label (vertically) and per class (horizontally) |
| D | Classification matrix of classified objects (order may be different from A) |
In crossvalidation it may be desired that sets of corresponding objects (e.g. pixels from the same image) are all together in the training set or in the test set. This might be enabled by adding an additional labeling to the dataset A (see ADDLABELS) corresponding to the sets and running LOSO with the corresponding LABLISTNAME. Alternatively, the set labels may be supplied in the call. In SET_LABLIST a ranking of the used labels can be supplied that will be used in C. In case SET_LABLIST does not contain all set labels used in SET_LABELS LOSO will only test the set labels given in SET_LABLIST and thereby perform areduced crosvalidation.
The reported error E identical to E = sum(C)./classsizes(D)*getprior(A)'; By confmat(D) a confusion matrix can be visualised.
datasets, mappings, testc, confmat,
| PRTools contents |