train_class_knn — Creates the search trees for a k-NN classifier.
train_class_knn creates the search trees for a k-NN classifier.
It is possible to set the number of trees via the parameters GenParamName and GenParamValue by 'num_trees'. The default value for the number of search trees is 4. A higher number of trees improves the accuracy of the search, but increases the run time.
It is possible to add more samples after training using the operator add_sample_class_knn. The added data affects the classification only, if train_class_knn is called again.
Automatic feature normalization can be activated by setting 'normalization' in GenParamName and 'true' in GenParamValue. The feature vectors are normalized by normalizing each dimension seperately. For each dimension, the mean and standard deviation is calculated over the training samples. Every feature vector is normalized by subtracting the mean and dividing by the standard deviation of the individual dimension. This results in a normalization, where each dimension has zero mean and unit variance. If the standard deviation happens to be zero, only the mean is subtracted. Please note however, that a feature dimension with no standard deviation does not change the classification result and should be removed. Automatic feature normalization will change the stored training data, but the original data can be restored at any time by calling train_class_knn with 'normalization' set to 'false'. If normalization is used, the operator classify_class_knn interprets the input data as unnormalized and performs normalization internally as it has been defined in the last call to train_class_knn.
This operator modifies the state of the following input parameter:
Handle of the k-NN classifier.
Names of the generic parameters that can be adjusted for the k-NN classifier creation.
Default value: 
List of values: 'normalization', 'num_trees'
Values of the generic parameters that can be adjusted for the k-NN classifier creation.
Default value: 
Suggested values: 4, 'false', 'true'
If the parameters are valid, the operator train_class_knn returns the value 2 (H_MSG_TRUE). If necessary, an exception is raised.
Marius Muja, David G. Lowe: “Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration”; International Conference on Computer Vision Theory and Applications (VISAPP 09); 2009.