Decision Tree to Ruleset
Converts (a single) decision tree model to PMML RuleSet model and also to a table containing the rules in a textual form. The resulting rules are independent of each other, the order of rules is not specified, can be changed. Missing value strategies are ignored, it will always evaluate to missing value.
Dialog Options
- Split rules to condition and outcome columns
- When checked, two columns will be created for the rules, Condition and Outcome, otherwise the rules will reside in a single column, Rule.
- Add confidence and weight columns
- From PMML the confidence and weight attributes are extracted to columns. (It will create columns with missing values.)
- Add Record count and Number of correct statistics columns
- In PMML, the recordCount and the nbCorrect attributes provide statistics about the input (training/test/validation) data, with this option, this information can be extracted to the columns: Record count and Number of correct
- Use additional parentheses to document precedence rules
- If checked the output will contain additional parenthesis around rule parts to clearly document precedence. For instance, NOT is a stronger operator than AND than OR - using parenthesis improves readability. Checking this option does not change any of the rule logic.
Ports
Input Ports
0 |
A PMML Decision Tree model. |
Output Ports
0 |
The decision tree model represented as PMML RuleSets (with firstHit rule selection method). |
1 |
The table contains the rules' text (in single (Rule) or two columns (Condition, Outcome), the rule Confidence and Weight information and optionally the Record count (for how many rows did the ruleset matched when created) and Number of correct values where the outcome of the rule matched the expected label when the model was created. |
This node is contained in KNIME Java Snippet Nodes
provided by KNIME GmbH, Konstanz, Germany.