Automatic learning of weight settings for multi-objective models
A historical scenario and historical decisions made in the historical scenario are received. The historical decisions represent a set of decision variables of an objective function. A random set of decision variables having different values than the set of decision variables are generated. To determine a weight setting associated with multiple objectives of the objective function, a number of inequalities are built and solved with an assumption that, for an optimization that minimizes the objective function, the objective function having the set of random decision variables has a larger value than the objective function having the set of decision variables. The receiving, the generating and the building steps may be repeated to determine multiple sets of weight settings. The multiple sets of weight settings are searched to select a target weight setting for each of the multiple objectives. The target weight setting may be automatically and continuously learned.
1. A computer-implemented method of automatic learning of weight settings of a multi-objective optimization, comprising:
receiving, by one or more processors, a historical scenario and historical decisions made in the historical scenario, the historical decisions representing a set of decision variables of an objective function, the objective function comprising multiple objectives, each of the multiple objectives comprising one or more decision variables from the set of decision variables;
generating, by one or more of the processors, a random set of decision variables having different values than the set of decision variables;
building, by one or more of the processors, a number of inequalities and solving the inequalities to determine a weight setting associated with each of the multiple objectives, the inequalities built with an assumption that, for an optimization that minimizes the objective function, the objective function having the set of random decision variables has a larger value than the objective function having the set of decision variables;
repeating the receiving, the generating and the building steps, wherein multiple sets of weight settings are determined for the multiple objectives;
searching the multiple sets of weight settings to select a target weight setting for each of the multiple objectives;
self-learning automatically by one or more of the processors by continuously and automatically updating the target weight setting associated with each of the multiple objectives based on continuously receiving additional decisions made to additional scenarios and automatically executing the generating, the building and the searching;
based on the target weight setting, automatically selecting a node for order fulfillment; and
controlling order fulfillment at the selected node by controlling an automated storage warehouse system associated with the selected node.
2. The method of claim 1 , wherein the objective function optimizes an online fulfillment order.
3. The method of claim 1 , wherein the historical decisions comprise user decisions.
4. The method of claim 1 , wherein the searching comprises selecting a confidence interval and searching the multiple sets of weight settings around the confidence interval.
5. The method of claim 1 , wherein the searching comprises maximizing a sum over the objective function values of each the multiple weight settings to select the target weight setting for each of the multiple objectives.
6. The method of claim 1 , wherein the objective function is pre-constructed.
7. The method of claim 1 , wherein the multiple objectives comprises shipping cost markdown cost, and operational cost.
8. A system of automatic learning weight settings of a multi-objective optimization, comprising:
one or more storage devices;
one or more hardware processors coupled to one or more of the storage devices,
one or more of the hardware processors operable to receive a historical scenario and historical decisions made in the historical scenario from one or more of the storage devices, the historical decisions representing a set of decision variables of an objective function, the objective function comprising multiple objectives, each of the multiple objectives comprising one or more decision variables from the set of decision variables,
one or more of the hardware processors further operable to generate a random set of decision variables having different values than the set of decision variables,
one or more of the hardware processors further operable to build a number of inequalities and solving the inequalities to determine a weight setting associated with each of the multiple objectives, the inequalities built with an assumption that, for an optimization that minimizes the objective function, the objective function having the set of random decision variables has a larger value than the objective function having the set of decision variables,
one or more of the hardware processors further operable to repeat the receiving, the generating and the building steps, wherein multiple sets of weight settings are determined for the multiple objectives,
one or more of the hardware processors further operable to search the multiple sets of weight settings to select a target weight setting for each of the multiple objectives,
one or more of the hardware processors further operable to self-learning automatically by continuously and automatically update the target weight setting associated with each of the multiple objectives based on continuously receiving additional decisions made to additional scenarios and automatically executing the generating, the building and the searching steps,
based on the target weight setting, one or more of the hardware processors automatically selecting a node for order fulfillment; and
one or more of the hardware processors further operable to control order fulfillment at the selected node by controlling an automated storage warehouse system associated with the selected node.
9. The system of claim 8 , wherein the objective function optimizes an online fulfillment order.
10. The system of claim 8 , wherein the historical decisions comprise user decisions.
11. The system of claim 8 , wherein one or more of the processors searches by selecting a confidence interval and searching the multiple sets of weight settings around the confidence interval.
12. The system of claim 8 , wherein one or more of the processors searches by maximizing a sum over the objective function values of each the multiple weight settings to select the target weight setting for each of the multiple objectives.
13. The system of claim 8 , wherein the objective function is pre-constructed.
14. The system of claim 8 , wherein the multiple objectives comprises shipping cost markdown cost, and operational cost.
15. 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, the program instructions readable by a device to cause the device to perform a method comprising:
receiving a historical scenario and historical decisions made in the historical scenario, the historical decisions representing a set of decision variables of an objective function, the objective function comprising multiple objectives, each of the multiple objectives comprising one or more decision variables from the set of decision variables;
generating a random set of decision variables having different values than the set of decision variables;
building a number of inequalities and solving the inequalities to determine a weight setting associated with each of the multiple objectives, the inequalities built with an assumption that, for an optimization that minimizes the objective function, the objective function having the set of random decision variables has a larger value than the objective function having the set of decision variables;
repeating the receiving, the generating and the building steps, wherein multiple sets of weight settings are determined for the multiple objectives;
searching the multiple sets of weight settings to select a target weight setting for each of the multiple objectives; and
self-learning automatically by continuously and automatically updating the target weight setting associated with each of the multiple objectives based on continuously receiving additional decisions made to the additional scenario and automatically executing the generating, the building and the searching steps;
based on the target weight setting, automatically selecting a node for order fulfillment; and
controlling order fulfillment at the selected node by controlling an automated storage warehouse system associated with the selected node.
16. The computer program product of claim 15 , wherein the objective function optimizes an online fulfillment order.
17. The computer program product of claim 15 , wherein the historical decisions comprise user decisions.
18. The computer program product of claim 15 , wherein the searching comprises selecting a confidence interval and searching the multiple sets of weight settings around the confidence interval.
19. The computer program product of claim 15 , wherein the searching comprises maximizing a sum over the objective function values of each the multiple weight settings to select the target weight setting for each of the multiple objectives.
20. The computer program product of claim 15 , wherein the objective function is pre-constructed and the multiple objectives comprises shipping cost markdown cost, and operational cost.