Learning optimization constraints through knowledge based data generation

1 . A system for learning all or part of a mathematical optimization model, comprising:

a memory device to store human-specified constraint and/or objective definitions that have been programmed in a general-purpose programming language by a human user, wherein the human-specified constraint is human-generated, the human-specified constraint and/or objective definitions are not designed for use with a specific optimization engine; and

a processor to:

select the specific optimization engine to generate a solution to an optimization problem;

generate labeled data for a plurality of solutions to the optimization problem using the human-generated constraint and/or objective definitions, wherein the processor in generating the labeled data is further configured to:

select values for each one of a plurality of decision variables;

input the values to an objective function to compute a value of the objective function; and

input the values to the human-generated constraint definitions to determine, for each solution, whether the values satisfy a corresponding constraints;

generate a formal constraint and/or objective model from the labeled data, wherein the formal constraint and/or objective model comprises automatically generated constraint and/or objective definitions that are syntactically different from the human-generated constraint and/or objective definitions and syntactically correct for the specific optimization engine; and

generate a formal objective model from the labeled data, wherein the formal objective model comprises an automatically generated objective function that is syntactically different from a human-generated objective function and syntactically correct for the specific optimization engine.

2 . The system of claim 1 , wherein the processor is further configured to generate a solution to the optimization problem using the automatically generated constraint and/or objective definitions, and execute a computer-readable instruction based on the solution.

3 . The system of claim 1 , wherein to select values of each one of the plurality of decision variables comprises to select the values, via an active learning algorithm, based on the labeled data generated for a previously processed solution or previous learning iterations.

4 . The system of claim 1 , wherein the labeled data comprises, for each solution:

real values of each decision variable input to an objective function of the optimization problem; and

a Boolean value for each solution indicating whether the values of the plurality of decision variables satisfy corresponding human-generated constraint definitions.

5 . The system of claim 4 , wherein the labeled data further comprises a value of the objective function(s) computed based on the input values of each solution.

6 . The system of claim 1 , wherein the automatically generated constraint/objective definitions are generated by a machine learning algorithm.

7 . A method of learning all or part of a mathematical optimization model, comprising:

receiving human-generated constraint and/or objective definitions that have been programmed in a general-purpose programming language by a human user, wherein the human-specified constraint and/or objective definitions are not designed for use with a specific optimization engine;

selecting the specific optimization engine to generate a solution to an optimization problem;

generating labeled data for a plurality of solutions to the optimization problem using the human generated constraint and/or objective definitions, wherein the generating the labeled data further comprises:

selecting values for each one of a plurality of decision variables;

inputting the values to an objective function to compute a value of the objective function; and

inputting the values to the human-generated constraint definitions to determine, for each solution, whether the values satisfy a corresponding constraints;

generating a formal constraint and/or objective model from the labeled data, wherein the formal constraint and/or objective model comprises automatically generated constraint and/or objective definitions that are syntactically different from the human-generated constraint and/or objective definitions and syntactically correct for the specific optimization engine; and

generating a formal objective model from the labeled data, wherein the formal objective model comprises an automatically generated objective function that is syntactically different from a human-generated objective function and syntactically correct for a specific optimization engine.

8 . The method of claim 7 , comprising generating a solution to the optimization problem using the automatically generated constraint and/or objective definitions, and executing a computer-readable instruction based on the solution.

9 . The method of claim 7 , wherein to select values of each one of the plurality of decision variables comprises to select the values, via an active learning algorithm, based on the labeled data generated for a previously processed solution or previous learning iterations.

10 . The method of claim 7 , wherein the labeled data comprises, for each solution:

real values of each solution variable input to an objective function of the optimization problem; and

a Boolean value for each solution indicating whether the values of the plurality of decision variables satisfy corresponding human-generated constraint definitions.

11 . The method of claim 10 , wherein the labeled data further comprises a value of the objective function(s) computed based on the input values of each solution variable.

12 . The method of claim 7 , wherein the automatically generated constraint/objective definitions are generated by a machine learning algorithm.

13 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, wherein the computer readable storage medium is not a transitory signal per se, and wherein the program instructions are executable by a processor to cause the processor to:

receive human-generated constraint and/or objective definitions that have been programmed in a general-purpose programming language by a human userthe human-specified constraint and/or objective definitions are not designed for use with a specific optimization engine;

select the specific optimization engine to generate a solution to an optimization problem;

generate labeled data for a plurality of solutions to the optimization problem using the human generated constraint and/or objective definitions, wherein the generating the labeled data is further configured to cause the processor to:

select values for each one of a plurality of decision variables;

input the values to an objective function to compute a value of the objective function; and

input the values to the human-generated constraint definitions to determine, for each solution, whether the values satisfy a corresponding constraints;

generate a formal constraint and/or objective model from the labeled data, wherein the formal constraint and/or objective model comprises automatically generated constraint and/or objective definitions that are syntactically different from the human-generated constraint and/or objective definitions and syntactically correct for the specific optimization engine; and

generate a formal objective model from the labeled data, wherein the formal objective model comprises an automatically generated objective function that is syntactically different from a human-generated objective function and syntactically correct for a specific optimization engine.

14 . The computer program product of claim 13 , wherein the labeled data comprises, for each solution:

real values of each decision variable input to an objective function of the optimization problem; and

a Boolean value for each solution indicating whether the values of the plurality of decision variables satisfy corresponding human-generated constraint definitions.