F_VN_CreateKnnModel

Syntax

Definition:

FUNCTION F_VN_CreateKnnModel : HRESULT
VAR_INPUT
    ipMlModel : Reference To  ITcVnMlModel;
    eKnnType  : ETcVnKnn;
    nK        : UDINT;
    hrPrev    : HRESULT;
END_VAR

 Inputs

 Return value

HRESULT

Further information

The function F_VN_CreateKnnModel creates a K-Nearest Neighbor (KNN) model.

K-Nearest Neighbor models

When training K-Nearest Neighbor models, all training samples are stored 1:1 in the model. The idea of data points within a multidimensional space (corresponding to the dimensionality of the samples) helps to understand this. A number nK of data points that are closest to the input sample (Euclidean distance) are used for prediction. These data points are called "nearest neighbors". All other data points are ignored in the prediction. The prediction result is calculated from all nearest neighbors, whereby these are weighted equally regardless of the actual distance to the input sample.

Parameter

Model type

eKnnType indicates whether the model is used as a classifier (TCVN_KNN_CLASSIFIER), as anomaly detection (TCVN_KNN_NOVELTY_DETECTOR) or as a regressor (TCVN_KNN_REGRESSOR).

Number of neighbors used

nK specifies how many data points are used as nearest neighbors in the prediction. For anomaly detection, nK is irrelevant because only the nearest neighbor is required. The value 1 should therefore be used in this case.

Application

For example, a K-nearest Neighbor model for classification using 5 nearest neighbors is created like this:

hr := F_VN_CreateKnnModel(
    ipMlModel   := ipMlModel,
    eKnnType    := TCVN_KNN_CLASSIFIER,
    nK          := 5,
    hrPrev      := hr);

Required License

TC3 Vision Machine Learning

System Requirements