QNet Approximator consists of four programs that compute lower and upper bounds on the optimal average cost for a class of stochastic processing networks by solving an approximate linear program (ALP). A fifth program generates policies from ALP solutions.
ALPs approximate the differential cost function as a linear combination of some set of basis functions. Supported basis functions include
- linear functions of x,
- quadratic functions of x,
- certain exponential of x,
- indicator functions for certain states x, and
- user-defined functions of x.
This software constructs linear programs which can be solved using CPLEX, if available, or with GLPK, a freely available open-source linear programming toolkit.
This software was developed with support from NSF Grant #CMMI-0620787.