set_params_class_knn — Set parameters for k-NN classification.
set_params_class_knn( : : KNNHandle, GenParamName, GenParamValue : )
set_params_class_knn sets parameters for the classification
of the k-nearest neighbors (k-NN) classifier KNNHandle.
It controls the behavior of classify_class_knn.
The value of 'k' can be set via GenParamName and
GenParamValue. Increasing 'k' also increases the accuracy
of the resulting neighbors and increases the run time.
The results can either be the determined class of the feature vector or
the indices of the nearest neighbors. The result behavior
can be selected with set_params_class_knn via the generic parameters
'method' and 'max_num_classes':
returns the nearest samples for each of maximally 'max_num_classes' different classes, if they have a representative in the nearest 'k' neighbors. The results are classes sorted by their minimal distance. There is no efficient way to determine in a k-NN-tree the nearest neighbor for exactly 'max_num_classes' classes.
counts the occurrences of certain classes among the nearest 'k' neighbors and returns the occurrent classes sorted by their relative frequency that is returned, too. Again, maximally 'max_num_classes' values are returned.
counts the occurrences of certain classes among the nearest 'k' neighbors and returns the occurrent classes sorted by their relative frequency weighted with the average distance that is returned, too. Again, maximally 'max_num_classes' values are returned.
returns the indices of the nearest 'k' neighbors and the distances.
The default behavior is 'classes_distance'.
The option 'num_checks' allows to set the number of maximal runs through the trees. The parameter has to be positive and the default value is 32. The higher this value is, the more accurate the results will be. As a trade-off, the running time will also be higher. Setting this parameter to 0 triggers an exact search.
The option 'epsilon' allows to set a stop criteria if the value is increased from the default value 0.0. The higher the value is set, the less accurate results of the estimated neighbors can be expected, while it might speed up the search.
This operator modifies the state of the following input parameter:
During execution of this operator, access to the value of this parameter must be synchronized if it is used across multiple threads.
KNNHandle (input_control, state is modified) class_knn → (handle)
Handle of the k-NN classifier.
GenParamName (input_control) string-array → (string)
Names of the generic parameters that can be adjusted for the k-NN classifier.
Default value: ['method','k','max_num_classes']
List of values: 'epsilon', 'k', 'max_num_classes', 'method', 'num_checks'
GenParamValue (input_control) number-array → (integer / real / string)
Values of the generic parameters that can be adjusted for the k-NN classifier.
Default value: ['classes_distance',5,1]
Suggested values: 'classes_distance', 'classes_frequency', 'classes_weighted_frequencies', 'neighbors_distance', 32, 0.0, 0.02, 0, 1, 2, 3, 4, 5, 6
If the parameters are valid, the operator set_params_class_knn
returns the value 2 (H_MSG_TRUE). If necessary, an exception is raised.
train_class_knn,
read_class_knn
create_class_knn,
read_class_knn,
get_params_class_knn
Marius Muja, David G. Lowe: “Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration”; International Conference on Computer Vision Theory and Applications (VISAPP 09); 2009.
Foundation