This release is a pre-release and may not be stable for production use.
ommx-openjij-adapter
Provides an adapter between OMMX and OpenJij.
Usage
ommx-openjij-adapter can be installed from PyPI:
pip install ommx-openjij-adapter
[!WARNING] The meaning of
SampleSet.objectiveschanged in OMMX v3. In v2, whenOMMXOpenJijSAAdapter.sample()applied a penalty method,SampleSet.objectivesincluded the penalty terms from the QUBO/HUBO energy (with the source sense restored for maximization). In v3 and later,SampleSet.objectivescontains the original problem's objective value. Penalty terms still affect OpenJij's sampling but are not included in the returned objective. Update code that treatedSampleSet.objectivesas the OpenJij energy.
The easy API prepares an isolated copy with the Adapter's recommended policy,
so it can directly accept inputs such as an Integer maximization model without
modifying the caller's Instance:
sample_set = OMMXOpenJijSAAdapter.sample(instance, num_reads=16)
OpenJij's initial_state is defined over the exact Adapter-input solver
variables; dictionary keys are those variable IDs. Because Preparation may
replace source variables with encoded or slack variables, pass initial_state
only to the exact-input constructor, sample_without_preparation(), or
solve_without_preparation() after preparing the Instance explicitly.
Preparation choices that require application knowledge remain explicit. For example, prepare the Instance with a selected fixed penalty weight and use the preparation-free API for a constrained model:
from ommx import DecisionVariable, FixedPenaltyPreparation, Instance
from ommx_openjij_adapter import OMMXOpenJijSAAdapter
x = DecisionVariable.binary(0, name="x")
instance = Instance.from_components(
decision_variables=[x],
objective=x,
constraints={0: x == 1},
sense=Instance.MINIMIZE,
)
policy = OMMXOpenJijSAAdapter.recommended_preparation_policy()
policy.fixed_penalty = FixedPenaltyPreparation.uniform_penalty_method_with_fixed_weight(
weight=2.0
)
instance.prepare(OMMXOpenJijSAAdapter.INPUT_CLASS, policy)
sample_set = OMMXOpenJijSAAdapter.sample_without_preparation(
instance,
num_reads=16,
)
print(sample_set.summary)
The fixed penalty weight is part of the preparation policy, not an OpenJij sampler parameter. The adapter deliberately does not choose a magnitude: zero adds no preference for feasibility, while a sufficiently large positive value is application-specific and still does not guarantee that every returned sample is feasible.
Accepted models and recommended preparation
This adapter directly accepts:
- Binary decision variables
- a polynomial objective of any degree (QUBO or Binary HUBO)
- no active regular or special constraints
- minimization
For the shared semantics of adapter input classes and preparation, see Adapter Input Classes and Explicit Constraint Lowering.
When building OpenJij sampler input, used variable IDs must fit a signed 64-bit integer and converted interaction coefficients must be finite. Violations are reported as conversion errors.
recommended_preparation_policy() returns a fresh editable
ommx.PreparationPolicy. It recommends:
- lowering active Indicator, OneHot, and SOS1 constraints;
- converting the active objective from maximization to minimization;
- attempting exact Integer slack with range 32, while permitting inequality-preserving Integer slack with upper bound 32 when exact equality conversion is unavailable; and
- log-encoding all used Integer variables.
Fixed penalty remains disabled until the caller selects either a uniform magnitude or magnitudes keyed by active regular constraint ID. Uniform weights are usually the convenient choice when special-constraint lowering creates regular constraints with generated IDs. OMMX validates the weight domain; the caller remains responsible for selecting a sufficient magnitude.
Pass the prepared Instance to sample_without_preparation. It remains the evaluation
owner for the returned SampleSet and retains the data needed to restore
source-variable values and evaluate removed constraints. The easy sample
API performs the copy and recommended preparation automatically when no custom
preparation choices are needed.
Integer log encoding follows the representability and bit-count limits of the
OMMX encoding operation. OMMX does not yet implement Kind::Spin; direct
OpenJij Spin input is tracked separately in
OMMX issue #1082.
1 maintainer
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
File details
Details for the file ommx_openjij_adapter-3.0.0b4.tar.gz.
File metadata
- Download URL: ommx_openjij_adapter-3.0.0b4.tar.gz
- Upload date:
- Size: 14.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via:
twine/7.0.0 CPython/3.13.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
b7468fa52bb5d9c7595282f4e83e260726335759857323968c3811ae39cc97a0
|
|
| MD5 |
e9c8edd8d92aa4ae553bc40e14cbe318
|
|
| BLAKE2b-256 |
43d63449e325ac04ea3e75577f03e621cc0ce54986898aa1d7e127accc2f2efd
|
Provenance
The following attestation bundles were made for ommx_openjij_adapter-3.0.0b4.tar.gz:
Publisher:
release_python.yml on Jij-Inc/ommx
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
ommx_openjij_adapter-3.0.0b4.tar.gz -
Subject digest:
b7468fa52bb5d9c7595282f4e83e260726335759857323968c3811ae39cc97a0 - Sigstore transparency entry: 2583202573
- Sigstore integration time:
-
Permalink:
Jij-Inc/ommx@d3c269a0fedf88d8b66c06eb7282904bf5709a9a -
Branch / Tag:
refs/tags/python-3.0.0b4 - Owner: https://github.com/Jij-Inc
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
release_python.yml@d3c269a0fedf88d8b66c06eb7282904bf5709a9a -
Trigger Event:
push
-
Statement type:
