I am using the HiGHS persistent solver to solve a two-stage stochastic optimization problem in a rolling-horizon framework. The persistent model is created once, and the model parameters are updated before each solve.
For each rolling-horizon run, I need to update the scenario probabilities. However, it appears that mpi-sppy embeds the scenario probabilities as constants in the Extensive Form (EF) objective when the EF is constructed, rather than representing them as mutable Pyomo parameters. As a result, changing the scenario probabilities between solves does not seem possible without rebuilding the EF.
Is there a recommended way to update scenario probabilities dynamically while reusing the same persistent model, or does the EF need to be reconstructed whenever the probabilities change?
I am using the HiGHS persistent solver to solve a two-stage stochastic optimization problem in a rolling-horizon framework. The persistent model is created once, and the model parameters are updated before each solve.
For each rolling-horizon run, I need to update the scenario probabilities. However, it appears that mpi-sppy embeds the scenario probabilities as constants in the Extensive Form (EF) objective when the EF is constructed, rather than representing them as mutable Pyomo parameters. As a result, changing the scenario probabilities between solves does not seem possible without rebuilding the EF.
Is there a recommended way to update scenario probabilities dynamically while reusing the same persistent model, or does the EF need to be reconstructed whenever the probabilities change?