Build your own model in Python

CGE-Core does not require model authors to inherit from a framework class.

Create a normal Python module:

from pyomo.environ import ConcreteModel, Constraint, Param, Set, Var

model_name = "MyCGE"
benchmark_only = {"sam0"}
shockable = {"tax", "endowment"}


def build_model(data):
    m = ConcreteModel()
    # declare sets, data, parameters, variables and equations
    return m


def apply_default_closure(model):
    # explicitly fix the numeraire / exogenous closure variables
    # and deactivate the model's redundant equation if applicable
    ...

Use it:

from cge_core.experimental.authoring import model_from_module

economy = model_from_module("my_model.py", data=my_data)
base = economy.solve()
scenario = base.scenario("Policy")
scenario.set("tax", "AGR", 0.0)
result = scenario.solve()

The author learns sets, calibration, variables, equations, closure, and shockability. Classes, inheritance, factories, decorators, and engine state are framework concerns, not modelling prerequisites.

See examples/custom_python_model/model.py for a complete tiny equilibrium.

Shockability declaration

shockable is optional. If it is omitted, the adapter remains open for model-author experimentation (subject to immutable-parameter and benchmark-only protections). If it is declared, it is an allow-list. In particular, shockable = set() intentionally means that nothing is shockable.

For temporary datasets created by StandardCGE.from_sam(), the returned economy also supports close() and the context-manager protocol when you want deterministic cleanup.