Coker

Coker is a mathematical programming toolkit and compiler pipeline for technical computing in Python. You define computations as ordinary Python callables, compile them into a symbolic representation, and lower that representation to multiple execution backends. The project is aimed at numerical modelling, optimisation, dynamics, and embedded-oriented execution workflows.

The current package metadata marks Coker as alpha software (Development Status :: 3 - Alpha).

What Coker does

Coker combines a few layers that usually live in separate tools:

• Backend lowering to numpy, casadi, jax, pytorch, and sympy; the separately packaged native coker backend can be registered for compact workspace-oriented execution.
• Differentiable execution models that support evaluation, composition, and conditional expressions.
• Dynamics and variational problem tooling for ODE systems, transcription helpers, and solver-backed optimisation workflows.
• Domain toolkits for spatial algebra, rigid-body kinematics, system modelling, and codesign-style mathematical programs.

Core capabilities

1. Compile Python callables into reusable functions

The primary entry point is coker.function. You describe argument spaces explicitly, provide a Python implementation, and choose a backend.

import numpy as np
from coker import function, Scalar, VectorSpace

f = function(
    arguments=[Scalar("x")],
    implementation=lambda x: 2 * x + 1,
    backend="numpy",
)
print(f(3))  # 7

A = np.array([[1.0, 0.0], [0.0, -1.0]])
g = function(
    arguments=[VectorSpace("x", 2)],
    implementation=lambda x: A @ x,
    backend="numpy",
)
print(g(np.array([1.0, 2.0])))  # [ 1. -2.]

2. Swap execution backends without rewriting the model

The same traced function can be lowered to different execution backends depending on the job:

• numpy for direct numerical execution
• casadi for optimisation-oriented symbolic workflows
• jax for alternate array execution
• pytorch for tensor-valued execution, autograd, and optional ODE/quadrature integration
• sympy for symbolic inspection and printing
• coker when the separately packaged coker_backend plugin is installed

from coker import function, Scalar

f_casadi = function(
    arguments=[Scalar("x")],
    implementation=lambda x: x**2,
    backend="casadi",
)

Optional extras declared by the package:

pip install "coker[casadi]"
pip install "coker[jax]"
pip install "coker[pytorch]"

Base installation:

pip install coker

Toolkit areas in this repository

Symbolic algebra and function composition

src/coker/algebra/ contains the tracing and function model used throughout the project. The test suite exercises:

• scalar and vector symbolic ops
• higher-order composition with FunctionSpace
• conditional expressions via if_then_else
• backend-specific lowering paths

Optional native Coker backend

The native coker backend is supplied by the separate coker_backend package. It lowers traced functions into a compact workspace-oriented graph; see docs/backend_architecture.rst for its artifact and runtime boundary.

Dynamics and optimisation

src/coker/dynamics/ exposes:

• create_autonomous_ode
• direct_sum
• VariationalProblem
• transcription helpers such as Legendre/LGR utilities
• backend-specific solver parameters and solve status reporting

The dynamics tests cover variational solvers, callbacks, direct-sum composition, and constrained parameter-fitting style problems.

Robotics and modelling toolkits

The repository also includes domain-focused toolkits under src/coker/toolkits/:

• spatial: rotations, isometries, screws, adjoint operators, quaternions
• kinematics: rigid-body trees, joints, inertias, forward kinematics, dynamics examples
• system_modelling: block/component modelling with a discoverable standard library
• codesign: a small problem-builder API for optimisation-style programs

The test suite includes concrete examples such as a single pendulum, double pendulum, SCARA manipulator, and hexapod leg models.

Repository layout

src/coker/        Python package source
examples/         Small runnable examples
scripts/          Standalone modelling scripts
tests/            Backend, symbolic, dynamics, and toolkit coverage
docs/             Sphinx documentation

Example files worth reading first

• docs/getting_started.rst — minimal symbolic function workflow
• docs/backend_architecture.rst — native backend execution model
• examples/pid_example.py — block-model style PID/plant composition
• scripts/double_pendulum.py — dynamics-oriented script example
• tests/benchmarks/benchmark_backends.py — benchmark scenarios for backend evaluation, ODE integration, and variational problems

Development

The repository uses uv in CI for environment management.

Install a development environment:

uv sync --group dev --extra casadi --extra jax --extra pytorch

Run the test suite:

uv run pytest

Build the documentation:

uv sync --group docs
uv run sphinx-build docs/ docs/_build/html -W --keep-going

Formatting and linting used in CI:

uv run black --check --diff src
uv run flake8 src tests docs examples scripts

Project status

The package metadata marks Coker as alpha-stage software. The repository already includes automated coverage for symbolic operations, backend lowering, dynamics, and toolkit examples.

License

Coker is licensed under the MPL-2.0. See LICENSE.TXT for the full text.