Converting Structures to Datasets

The Convert Structure to Dataset task produces a dataset that contains all the information included in the specified structure. 


There are many reasons why you may want to convert structures, here are a few examples:

  • in-depth analysis can be performed on the resulting dataset through a Data Manager. For example you could filter rules through a formula to identify only those rules which have a specific range of covering.

  • the resulting dataset can be exported to an external item, such as a file or database table

  • produce a dataset that contains all the information included in a model, such as the weights of a neural networks or the coefficients of a linear regression.


The more specific Convert Ruleset to Dataset (and vice-versa) and Convert Model to Dataset (and vice-versa) tasks are still available for backward compatibility, but they are fully substituted by this new set of generic structure conversion tasks.


Prerequisites


Procedure

  1. Drag and drop the Convert Structure to Dataset task onto the stage.

  2. Connect a task that contains an existing model to the Convert Structure to Dataset task.

  3. Double click the task and select the required structure. The only structure that requires additional parameters are rules, which require the parameters explained in the table below.

  4. Save and compute the task. Care must be taken when modifying the dataset provided by this task, since this can lead to an inconsistent format which can no longer be interpreted by a subsequent Convert Dataset to Structure task.

Convert Structure to Dataset options for Rulesets

Parameter Name

PO

Description

Dataset format

rulewide

Select the required format for the dataset (rulewide), which can be one of the following:

  • One row for each term: the resulting table contains a row for each condition attribute value within each rule

  • One row for each condition: the resulting table contains a row for each condition attribute within each rule

  • One rule for each rule: the resulting table contains a row for each rule

Conditions on ordered attributes format

ruleinterval

Select the required format for the conditions on ordered attributes (ruleinterval), which can be one of the following:

  • a < x < b: values will always be displayed with greater than or less than indicators

  • x > a, x in [a,b]: values will be displayed with greater than or less than indicators, or as a range when possible

  • x in [0,Inf], x in [a,b]: values will always be displayed as a range, using an infinite value (Inf) when these is no end of range value available.