Initial Values by Name and Subscripts

SCIP accepts initial values grouped by variable name and tuple subscripts. The adapter resolves them against the instance being solved, so the same input can be reused when rebuilding a model changes variable IDs.

from ommx import DecisionVariable, Instance, NamedState
from ommx_pyscipopt_adapter import OMMXPySCIPOptAdapter

x = DecisionVariable.integer(31, lower=0, upper=5, name="x", subscripts=[0])
y = DecisionVariable.integer(107, lower=0, upper=5, name="x", subscripts=[1])
instance = Instance.from_components(
    decision_variables=[x, y], objective=2 * x + y,
    constraints=[x + y <= 5], sense=Instance.MAXIMIZE,
)
initial_values: NamedState = {"x": {(0,): 1.0, (1,): 4.0}}
solution = OMMXPySCIPOptAdapter.solve(instance, initial_state=initial_values)
assert solution.state.entries == {x.id: 5.0, y.id: 0.0}

To obtain a State explicitly, use Instance.state_from_named_values():

state = instance.state_from_named_values(initial_values)
assert state.entries == {x.id: 1.0, y.id: 4.0}

import math

assert all(math.isfinite(value) for value in state.entries.values())
assert instance.evaluate(state).feasible

Use tuple keys: () for scalars, (i,) for one index, and (i, j) for two. The output of Solution.extract_all_decision_variables() can also be converted against another instance’s IDs.

Unknown name/subscript pairs raise KeyError; ambiguous pairs raise ValueError. Parameters and descriptions do not distinguish keys. For ambiguous variables, use an ID-keyed state. Partial mappings are accepted and omitted values are not filled. SCIP initializes omitted solver variables to zero. Keys that collide after string or integer normalization raise ValueError. Conversion does not check finiteness, bounds, kind, fixed values, or dependencies. To require finite values, explicitly check the converted state’s entries with math.isfinite; evaluation in SDK 2.x does not guarantee finiteness. Evaluate a complete state with Instance.evaluate() and inspect the returned solution.feasible for variable-kind, bound, and constraint feasibility. Evaluation can return an infeasible solution and also resolves fixed values and dependencies. Neither the instance nor the input is modified by conversion.