System and method for automatic parameter tuning of campaign planning with hierarchical linear programming objectives
A system and method are disclosed for big bucket campaign planning that automatically learns weights for parameters of a weighted evaluation function. Embodiments include modeling the use of the one or more campaign operations and one or more campaignable resources as one or more campaign planning problems comprising a sequential decision problem with decision parameters, determining a campaign plan for the use of campaign operations and one or more campaignable resource, encoding a required policy of the sequential decision problem into a k-lookahead search strategy by defining an evaluation function comprising a weighted sum of features evaluated from the campaign planning problem, learning weights associated with an evaluation function that determines an effective objective function for campaign planning as a linear programming problem, and computing the parameters of the evaluation function using an iterative cross-entropy campaign planning.
1. A system of big bucket campaign planning that automatically learns weights for parameters of a weighted evaluation function, comprising:
a computer, comprising a processor and a memory, configured to:
model the use of the one or more campaign operations and one or more campaignable resources of a supply chain network comprising a production line configured to produce one or more products using one or more campaign operations and one or more campaignable resources as one or more campaign planning problems comprising a sequential decision problem with one or more decision parameters;
define an evaluation function comprising a weighted sum of features evaluated from the campaign planning problem that encodes a required policy;
initialize weights to build a consumption profile and evaluation function;
generate random weight vector samples to determine fitness values, wherein the fitness values indicate a level of variability:
evaluate reward values based on the fitness values associated with the random weight vector samples;
select a sub-sample of the top fitness values having the best associated objective function;
repeat the generate, the evaluate and the select steps to adjust the weights until a stopping criterion is met indicating an optimal solution has been reached regarding the modelled campaign operations and the modelled campaignable resources;
determine a campaign plan for the use of the one or more campaign operations and one or more campaignable resource.
2. The system of claim 1 , wherein the computer is further configured to:
invoke a user-specified number of processes to perform parallel operations of a campaign planning solve, each of the parallel operations running for a given weight sample and remaining independent of the other parallel operations.
3. The system of claim 2 , wherein the evaluation function comprises a weighted consumption profile that is evaluated according to one or more measures of on-hand inventory, demand, and safety stock in a predetermined number of future buckets.
4. The system of claim 3 , wherein the sequential decision problem decision parameters comprise one or more of the value function, a reward function, a state space, and an action set.
5. The system of claim 4 wherein the evaluation function comprises a linear combinations of the features (f1, f2, f3, . . . , fd) weighted by coefficients (w1, w2, w3, . . . , wd).
6. A computer-implemented method of big bucket campaign planning that automatically learns weights for parameters of a weighted evaluation function, comprising:
modeling, by a computer, the use of the one or more campaign operations and one or more campaignable resources of a supply chain network comprising a production line configured to produce one or more products using one or more campaign operations and one or more campaignable resources as one or more campaign planning problems comprising a sequential decision problem with one or more decision parameters;
defining an evaluation function comprising a weighted sum of features evaluated from the campaign planning problem that encodes a required policy;
initializing weights to build a consumption profile and evaluation function;
generating random weight vector samples to determine fitness values, wherein the fitness values indicate a level of variability;
evaluating reward values based on the fitness values associated with the random weight vector samples;
selecting a sub-sample of the top fitness values having the best associated objective function;
repeating the generating, the evaluating and the selecting steps to adjust the weights until a stopping criterion is met indicating an optimal solution has been reached regarding the modelled campaign operations and the modelled campaignable resources;
determining, by the computer, a campaign plan for the use of the one or more campaign operations and one or more campaignable resource.
7. The computer-implemented method of claim 6 , further comprising:
invoking, by the computer, a user-specified number of processes to perform parallel operations of a campaign planning solve, each of the parallel operations running for a given weight sample and remaining independent of the other parallel operations.
8. The computer-implemented method of claim 7 , wherein the evaluation function comprises a weighted consumption profile that is evaluated according to one or more measures of on-hand inventory, demand, and safety stock in a predetermined number of future buckets.
9. The computer-implemented method of claim 8 , wherein the sequential decision problem decision parameters comprise one or more of the value function, a reward function, a state space, and an action set.
10. The computer-implemented method of claim 9 , wherein the evaluation function comprises a linear combinations of the features (f1, f2, f3, . . . , fd) weighted by coefficients (w1, w2, w3, . . . , wd).
11. A non-transitory computer-readable medium embodied with software for big bucket campaign planning that automatically learns weights for parameters of a weighted evaluation function software, the software when executed:
models the use of the one or more campaign operations and one or more campaignable resources of a supply chain network comprising a production line configured to produce one or Amendment more products using one or more campaign operations and one or more campaignable resources as one or more campaign planning problems comprising a sequential decision problem with one or more decision parameters;
defines an evaluation function comprising a weighted sum of features evaluated from the campaign planning problem that encodes a required policy;
initializes weights to build a consumption profile and evaluation function;
generates random weight vector samples to determine fitness values, wherein the fitness values indicate a level of variability;
evaluates reward values based on the fitness values associated with the random weight vector samples;
selects a sub-sample of the top fitness values having the best associated objective function;
repeats the generates, the evaluates and the selects steps to adjust the weights until a stopping criterion is met indicating an optimal solution has been reached regarding the modelled campaign operations and the modelled campaignable resources;
determines a campaign plan for the use of the one or more campaign operations and one or more campaignable resource.
12. The non-transitory computer-readable medium of claim 11 , further comprising:
invokes a user-specified number of processes to perform parallel operations of a campaign planning solve, each of the parallel operations running for a given weight sample and remaining independent of the other parallel operations.
13. The non-transitory computer-readable medium of claim 12 , wherein the evaluation function comprises a weighted consumption profile that is evaluated according to one or more measures of on-hand inventory, demand, and safety stock in a predetermined number of future buckets.
14. The non-transitory computer-readable medium of claim 13 , wherein the sequential decision problem decision parameters comprise one or more of the value function, a reward function, a state space, and an action set.