IDAX.PREDICT_KNN - Apply a KNN model

Use this stored procedure to apply a KNN model to generate classification predictions or regression predictions for a data set.

Authorization

The privileges held by the authorization ID of the statement must include the IDAX_USER role. Additionally, you must be the owner of the model or have the authority to alter it.

Syntax

IDAX.PREDICT_KNN(in parameter_string varchar(32672))

Parameter descriptions

parameter_string
Mandatory one-string parameter that contains pairs of <parameter>=<value> entries that are separated by a comma.
Data type: VARCHAR(32672)
The following list shows the parameter values:
model
Mandatory.
The name of the KNN model that is to be applied.
Data type: VARCHAR(64)
intable
Mandatory.
The name of the input table.
Data type: VARCHAR(128)
outtable
Mandatory.
The name of the output table where the predicted values for the records of the input table are stored.
Data type: VARCHAR(128)
id
Optional.
The column of the input table that identifies a unique record.
Default: The name of the id column that is used to build the model.
Data type: VARCHAR(128)
target
Optional.
The column of the input table that represents the prediction target.
The specified target column is not used for prediction.
Default: The name of the target column that is used to build the model.
Data type: VARCHAR(128)
distance
Optional.
The distance that is to be used for calculating the nearest neighbors.
Allowed values are "euclidean", "canberra", "manhattan", and "maximum".
Default: 'euclidean'
Data type: VARCHAR(16)
k
Optional.
The number of the nearest neighbors that are to be considered.
Default: 3
Range: >=1
Data type: INTEGER
stand
Optional.
A flag that indicates to standardize the continuous input columns of the input table and of the model table.
Different scales of the continuous input columns are ignored so that columns with bigger values are no longer disadvantaged.
Default: true
Data type: BOOLEAN
fast
Optional.
A flag that indicates whether the faster approximate calculation method is to be used instead of the exact method.
Default: true
Data type: BOOLEAN
weights
Optional.
A table that contains class weights for the model table.
Class weights are ignored if the target column of the model is of type "continuous".
If you do not specify this parameter, all the weights are assumed to be equal to 1.
The weight table contains the following columns:
  1. A WEIGHT column that contains numeric positive weights
  2. A CLASS column that contains values of the target column of the model
Data type: VARCHAR(128)

Returned information

The number of input sequences for which a prediction is made as result set.

Example

CALL IDAX.PREDICT_KNN('model=customer_censor_mdl, intable=customer_churn, outtable=customer_censor_score');