PySP

PySP, or Python Stochastic Programming, is a component of the Pyomo optimization framework designed for modeling and solving stochastic programming problems that are represented by scenario trees. It focuses on two-stage and multi-stage problems in which decisions are made under uncertainty and evaluated across multiple scenarios with given probabilities.

The package provides tools to define scenario trees, assign probabilities, and connect them to a Pyomo template model that expresses the deterministic subproblems for each scenario. The solver process typically involves a master problem for first-stage decisions and scenario-specific subproblems for later stages, unified by the probability-weighted objective and linking constraints.

PySP supports different solution approaches, including decomposition-based algorithms such as Progressive Hedging and the L-shaped method, as well as scenario reduction and sampling techniques to manage large numbers of scenarios. It can generate a single reformulated stochastic program or operate in a decomposed fashion during solving.

Usage: a user constructs a scenario tree (or a generator), writes a Pyomo model that defines decisions across stages, and uses PySP to solve by invoking a solver on the generated stochastic program. Result summaries include first-stage decisions and scenario-dependent decisions and costs.

Development and availability: PySP is open-source and maintained as part of the Pyomo project. It integrates with common optimizers such as CPLEX, Gurobi, and open-source solvers, and is documented on the Pyomo website and GitHub.