Amplify-BBOpt

Amplify-BBOpt is a powerful Python package by Fixstars Amplify, designed to streamline the implementation and execution of black-box optimization algorithms.

• Documentation:
  https://amplify.fixstars.com/docs/amplify-bbopt/v1/
• Contributing (bug reports, improvements):
  https://amplify.fixstars.com/en/contact

Features

Amplify-BBOpt is built upon the powerful Amplify SDK, bridging the gap between black-box optimization and specialized hardware. It uniquely offers annealing-based optimization methods, including Factorization Machine Quantum Annealing (FMQA) and Kernel-QA, which are designed to run on quantum annealers and various Ising machine backends.

• Intuitive API: A clean, straightforward interface enables rapid experimentation and allows users to easily swap between different optimization methods.
• High-Dimensional Scalability: Engineered to effectively solve complex optimization problems, even those with a high-dimensional parameter space.
• Built-in Constraint Handling: Natively supports both unconstrained and constrained optimization problems, providing a versatile framework for real-world applications.

Basic Usage

from amplify import AmplifyAEClient

from amplify_bbopt import FMTrainer, Optimizer, RealVariable, blackbox


# Define a test function
# (This represents a simulation or experiment)
def spherical_func(a: float, b: float) -> float:
    return (a + 1) ** 2 + (b - 1) ** 2


# Make a black-box function and define relevant decision variables
@blackbox
def bb_func(
    a: float = RealVariable((-5, 5)),  # type: ignore
    b: float = RealVariable((-5, 5)),  # type: ignore
) -> float:
    return spherical_func(a, b)


# Prepare solver client
client = AmplifyAEClient()
client.parameters.time_limit_ms = 1000

# Execute optimization
opt = Optimizer(bb_func, FMTrainer(), client)
opt.add_random_training_data(num_data=5)
opt.optimize(num_iterations=10)

# View results
print(opt.best.objective)
print(opt.best.values)

For Developers

Amplify-BBOpt requires Python 3.10 or later. Also, it depends on amplify and torch packages, which can be installed via pip. After installing the dependencies, you can install Amplify-BBOpt in editable mode.

$ pip install -e amplify-bbopt

Instead, we recommend using uv for managing the development environment and running tests.

$ uv sync

Testing:

$ uv run pytest