discopt.stochastic.extensive_form

Extensive form (deterministic equivalent) of a two-stage stochastic program.

Builds the single monolithic model min c·x + risk_s[ Q_s(x) ] by calling a scenario-creator callback once per scenario (the mpi-sppy / PySP pattern): the callback constructs that scenario’s second-stage variables and constraints (given the realized data) and returns its recourse cost expression. The risk measure (risk) then assembles the objective.

This is correct by construction (an exact deterministic equivalent), so it is the Phase 0 baseline and the oracle the decomposed methods (L-shaped, PH) are checked against. See docs/dev/stochastic-module-plan.md §1, §3.

Classes

ExtensiveForm

Result of build_extensive_form() (kept for inspection/testing).

Functions

build_extensive_form(→ ExtensiveForm)

Assemble the deterministic-equivalent objective onto model (in place).

Module Contents

class discopt.stochastic.extensive_form.ExtensiveForm

Result of build_extensive_form() (kept for inspection/testing).

scenario_costs: list[discopt.modeling.core.Expression]

objective: discopt.modeling.core.Expression

risk: discopt.stochastic.risk.RiskMeasure

discopt.stochastic.extensive_form.build_extensive_form(model: discopt.modeling.core.Model, *, scenarios: discopt.stochastic.scenario.ScenarioSet, recourse_builder, first_stage_cost: discopt.modeling.core.Expression | None = None, risk: discopt.stochastic.risk.RiskMeasure | None = None, minimize: bool = True) → ExtensiveForm

Assemble the deterministic-equivalent objective onto model (in place).

Parameters:

• scenarios – The probability-weighted scenario set.

• recourse_builder – Callable (model, scenario_data, s) -> recourse_cost_expr. It must create scenario s’s second-stage variables (uniquely named, e.g. suffixed by s) and add its recourse constraints to model, returning the scenario’s recourse cost expression (to be minimized in expectation / risk).

• first_stage_cost – The here-and-now cost c·x (default 0).

• risk – A RiskMeasure (default Expectation).