Generating a Derive node

When you generate a Derive node, it applies the inverse target transformation to the score field. By default the node enters the name of the score field that would be produced by an automodeler (such as Auto Classifier or Auto Numeric) or the Ensemble node. If a scale (range) target has been transformed the score field is shown in transformed units; for example, log($) instead of $. In order to interpret and use the results, you must convert the predicted value back to the original scale.

Note: You can only generate a Derive node when the ADP node contains an analysis which rescales a range target (that is, Box-Cox rescaling is selected on the Prepare Inputs & Target panel). You cannot generate a Derive node if the target is not a range, or if the Box-Cox rescaling is not selected.

The Derive node is created in Multiple mode and uses @FIELD in the expression so you can add the transformed target if required. For example using the following details:

  • Target field name: response
  • Transformed target field name: response_transformed
  • Score field name: $XR-response_transformed

The Derive node would create a new field: $XR-response_transformed_inverse.

Note: If you are not using an automodeler or Ensemble node, you will need to edit the Derive node to transform the correct score field for your model.

Normalized continuous targets

By default, if you select the Rescale a continuous target with a Box-Cox transformation check box on the Prepare Inputs & Target panel this transforms the target and you create a new field that will be the target for your model building. For example if your original target was response, the new target will be response_transformed; models downstream of the ADP node will pick this new target up automatically.

However, this may cause issues, depending on the original target. For example, if the target was Age, the values of the new target will not be Years, but a transformed version of Years. This means you cannot look at the scores and interpret them since they aren’t in recognizable units. In this case you can apply an inverse transformation that will turn your transformed units back into whatever they were meant to be. To do this:

  1. After clicking Analyze Data to run the ADP analysis, select Derive Node from the Generate menu.
  2. Place the Derive node after your nugget on the model canvas.

The Derive node will restore the score field to the original dimensions so that the prediction will be in the original Years values.

By default the Derive node transforms the score field generated by an auto-modeler or an ensembled model. If you are building an individual model, you need to edit the Derive node to derive from your actual score field. If you want to evaluate your model, you should add the transformed target to the Derive From field in the Derive node. This applies the same inverse transformation to the target and any downstream Evaluation or Analysis node will use the transformed data correctly as long as you switch those nodes to use field names instead of metadata.

If you also want to restore the original name, you can use a Filter node to remove the original target field if it’s still there, and rename the target and score fields.