OMMX Python SDK 1.8.0

Please refer to the GitHub Release for individual changes.

⚠️ Includes breaking changes due to the addition of SolverAdapter.

Summary

• Added a new SolverAdapter abstract base class to serve as a common interface for adapters to different solvers.

• ommx-python-mip-adapter and ommx-pyscipopt-adapter have been changed to use SolverAdapter according to the adapter implementation guide

  • ⚠️ This is a breaking change. Code using these adapters will need to be updated.

  • Other adapters will be updated in future versions.

Solver Adapter

The introduction of the SolverAdapter base class aims to make the API for different adapters more consistent. ommx-python-mip-adapter and ommx-pyscipopt-adapter now use the SolverAdapter base class.

Here is an example of the new Adapter interface to simply solve an OMMX instance.

from ommx.v1 import Instance, DecisionVariable
from ommx_python_mip_adapter import OMMXPythonMIPAdapter

p = [10, 13, 18, 32, 7, 15]
w = [11, 15, 20, 35, 10, 33]
x = [DecisionVariable.binary(i) for i in range(6)]
instance = Instance.from_components(
    decision_variables=x,
    objective=sum(p[i] * x[i] for i in range(6)),
    constraints=[sum(w[i] * x[i] for i in range(6)) <= 47],
    sense=Instance.MAXIMIZE,
)

solution = OMMXPythonMIPAdapter.solve(instance)
solution.objective

42.0

With the new update, the process looks the same as the above when using the OMMXPySCIPOptAdapter class instead.

To replace the usage of instance_to_model() functions, you can instantiating an adapter and using solver_input. You can then apply any solver-specific parameters before optimizing manually, then calling decode() to obtain the OMMX solution.

adapter = OMMXPythonMIPAdapter(instance)
model = adapter.solver_input # in OMMXPythonMIPAdapter's case, this is a `mip.Model` object
# modify model parameters here
model.optimize() 
solution = adapter.decode(model)
solution.objective

42.0