Arguments of Trainable Unit Functions
The field types supported by the Trainable Unit functions are: Integer, Long, Float, Double, Boolean, String, and Date.
An Integer, Long, Float, or Double field that does not have selection values represents a numeric attribute.
For regression problems, the label field must be of one of the four numeric types. Note that irrespective of the type of the label field the predicted values will be of the Double type.
A field of any supported type (including the numeric ones) that has selection values, and also a field of the Boolean type, represents a nominal attribute.
For classification problems, the label column must be nominal. The best way to represent a nominal attribute or a nominal label field is to use a String field with selection values that contain all the possible classes. Note that irrespective of the type of the label field the predicted class will be of the String type (with selection values that list all possible classes).
If the task is specified as classification and the label field does not have selection values, the trainable unit will try to convert the label values into nominal ones using all unique values as selection values. However it is assumed that the test set will not contain any value that was not present in the training set (i.e. the test set should not contain a class that was not present in the training set).
A Date field represents a date attribute.
Null values are treated as missing values. Also, if a prediction for a certain data instance cannot be made the operate function will return Null as a prediction.
 | The format of the data table passed to the train function is saved as a reference for format validation. Hence, the data table passed to the Operate or the Evaluate function must have exactly the same format as the format of the data table used for training (including selection values, nullability, and other format options). If the formats do not match an error will occur. |
The data tables passed to the Train, Evaluate, or Cross Validate functions must contain at least one instance. The number of instances also cannot be smaller than the number of folds for cross validation.
Was this page helpful?