Methods and apparatus to improve decision tree execution
Methods, apparatus, systems and articles of manufacture are disclosed to improve decision tree execution. An example method includes retrieving, with a processor, a decision tree logic expression in a sum-of-products (SOP) form, the decision tree logic expression consuming a first duration to evaluate a dataset, eliminating, with the processor, redundant variables of the decision tree logic expression by transforming the decision tree logic expression into a product-of-sums (POS) form, and evaluating, with the processor, the data set with the decision tree logic expression in the POS form, the decision tree logic expression in the POS form consuming a second duration to evaluate the data set that is less than the first duration.
1. A method to reduce decision tree evaluation time, comprising:
retrieving, from a storage location via a network, by executing instructions with a processor, a decision tree logic expression in a sum-of-products (SOP) form, the SOP form including first subexpressions having a logical OR relationship, the decision tree logic expression consuming a first duration to evaluate a dataset;
eliminating, by executing instructions with the processor, redundant variables of the decision tree logic expression by transforming the decision tree logic expression into a product-of-sums (POS) form, the POS form including second subexpressions having a logical AND relationship; and
evaluating, by executing instructions with the processor, the data set with the decision tree logic expression in the POS form, the decision tree logic expression in the POS form consuming a second duration to evaluate the data set that is less than the first duration.
2. A method as defined in claim 1 , further including identifying a first variable that is common to at least two of the first subexpressions.
3. A method as defined in claim 2 , further including generating a simplified version of the transformed decision tree logic expression that factors out the first variable common to the at least two of the first subexpressions.
4. A method as defined in claim 1 , further including identifying surrogate variables associated with the transformed decision tree logic expression, the surrogate variables including a non-Boolean value.
5. A method as defined in claim 4 , further including replacing the non-Boolean value with a correction value when the non-Boolean value cannot be evaluated with an inequality test.
6. A method as defined in claim 5 , wherein the inequality test includes at least one of a greater-than inequality test, a less-than inequality test, or an equal test.
7. A method as defined in claim 5 , wherein the inequality test is associated with a threshold value.
8. A method as defined in claim 5 , wherein the non-Boolean value is at least one of NULL or missing.
9. A method as defined in claim 5 , further including reducing an evaluation computation burden on the processor by assigning the correction value as a binary value.
10. An apparatus to reduce decision tree evaluation time, comprising:
a decision tree interface to retrieve, from a storage location via a network, a decision tree logic expression in a sum-of-products (SOP) form, the SOP form including first subexpressions having a logical OR relationship, the decision tree logic expression consuming a first duration to evaluate a data set;
a tree transformation engine to eliminate redundant variables of the decision tree logic expression by transforming the decision tree logic expression into a product-of-sums (POS) form, the POS form including second subexpressions having a logical AND relationship; and
a segmentation optimizer to evaluate the data set with the decision tree logic expression in the POS form, the decision tree logic expression in the POS form consuming a second duration to evaluate the data set that is less than the first duration.
11. An apparatus as defined in claim 10 , further including a tree factorization engine to identify a first variable that is common to at least two of the first sub expressions.
12. An apparatus as defined in claim 11 , wherein the tree factorization engine is to generate a simplified version of the transformed decision tree logic expression that factors out the first variable common to the at least two of the first subexpressions.
13. An apparatus as defined in claim 10 , further including a variable evaluator to identify surrogate variables associated with the transformed decision tree logic expression, the surrogate variables including a non-Boolean value.
14. An apparatus as defined in claim 13 , further including a not-a-number (NaN) computation engine to replace the non-Boolean value with a correction value when the non-Boolean value cannot be evaluated with an inequality test.
15. An apparatus as defined in claim 14 , wherein the inequality test includes at least one of a greater-than inequality test, a less-than inequality test, or an equal test.
16. An apparatus as defined in claim 14 , wherein the inequality test is associated with a threshold value.
17. An apparatus as defined in claim 14 , wherein the non-Boolean value is at least one of NULL or missing.
18. An apparatus as defined in claim 14 , wherein the NaN computation engine is to reduce an evaluation computation burden on a processor by assigning the correction value as a binary value.
19. A tangible computer readable storage medium comprising computer readable instructions that, when executed, cause a processor to at least:
retrieve, from a storage location via a network, a decision tree logic expression in a sum-of-products (SOP) form, the SOP form including first subexpressions having a logical OR relationship, the decision tree logic expression consuming a first duration to evaluate a dataset;
eliminate redundant variables of the decision tree logic expression by transforming the decision tree logic expression into a product-of-sums (POS) form, the POS form including second subexpressions having a logical AND relationship; and
evaluate the data set with the decision tree logic expression in the POS form, the decision tree logic expression in the POS form consuming a second duration to evaluate the data set that is less than the first duration.
20. A tangible computer readable storage medium as defined in claim 19 , wherein the instructions, when executed, cause the processor to identify a first variable that is common to at least two of the first subexpressions.
21. A tangible computer readable storage medium as defined in claim 20 , wherein the instructions, when executed, cause the processor to generate a simplified version of the transformed decision tree logic expression that factors out the first variable common to the at least two of the first subexpressions.
22. A tangible computer readable storage medium as defined in claim 19 , wherein the instructions, when executed, cause the processor to identify surrogate variables associated with the transformed decision tree logic expression, the surrogate variables including a non-Boolean value.
23. A tangible computer readable storage medium as defined in claim 22 , wherein the instructions, when executed, cause the processor to replace the non-Boolean value with a correction value when the non-Boolean value cannot be evaluated with an inequality test.
24. A tangible computer readable storage medium as defined in claim 23 , wherein the instructions, when executed, cause the processor to reduce an evaluation computation burden on the processor by assigning the correction value as a binary value.
25. A method as defined in claim 1 , further including determining branch occurrence rates for the decision tree logic expression in the POS form.
26. A method as defined in claim 25 , further including ranking branches of the decision tree logic expression in the POS form by the branch occurrence rates.
27. A method as defined in claim 26 , further including rearranging the branches of the decision tree logic expression in the POS form based on the ranking of the branches.