GAlibrate is a python toolkit that provides an easy to use interface for model calibration/parameter estimation using an implementation of continuous genetic algorithm-based optimization. Its functionality and API were designed to be familiar to users of the PyDREAM, simplePSO, and Gleipnir packages.
Although GAlibrate provides a general framework for running continuous genetic algorithm-based optimizations, it was created with systems biology models in mind. It therefore supplies additional tools for working with biological models in the PySB format.
- Julia integration - New version of core GA that ports some key funtions to Julia using the PyJulia package.
- New
benchmarksmodule defining a set of functions used to benchmark and test single objective optimazation routines. - Test suite using pytest with 63% overall coverage.
- Updated profiling and performance benchmarking Jupyter notebooks.
- Function to resume/continue GAO runs for additional generations:
GAO.resume. - Several new example cases under examples
- core GA now returns an array with fitness value of the fittest individual from each generation which can be accessed from the GAO property
GAO.best_fitness_per_generation. - Bug fix in core GA for sorting the population before selection and mating.
- Optional progress bar to monitor passage of generations during GAO run that is only displayed if tqdm is installed
- Optional multiprocessing based parallelism when evaluating the fitness function over the population during a GAO run.
| ! Note |
|---|
| GAlibrate is still in version zero development so new versions may not be backwards compatible. |
GAlibrate installs as the galibrate package. It is compatible (i.e., tested) with Python 3.10.11.
Note that galibrate has the following core dependencies:
You can install the latest release of the galibrate package using pip sourced from the GitHub repo -
Fresh install:
pip install https://github.com/blakeaw/GAlibrate/archive/refs/tags/v0.7.2.zip
Or to upgrade from an older version:
pip install --upgrade https://github.com/blakeaw/GAlibrate/archive/refs/tags/v0.7.2.zip
galibrate can also be pip installed from PyPI,
pip install galibrate
but this version currently doesn't include the Cython accelerated version of the core GA algorithm.
You can install the galibrate package from the blakeaw channel:
conda install -c blakeaw galibrate
NumPy and SciPy dependencies will be automatically installed with this version.
The following software is not required for the basic operation of GAlibrate, but provides extra capabilities and features when installed.
GAlibrate includes an implementation of the core genetic algorithm that is written in Cython, which takes advantage of Cython-based optimizations and compilation to accelerate the algorithm. This version of genetic algorithm is used if Cython is installed.
GAlibrate also includes an implementation of the core genetic algorithm that takes advantage of Numba-based JIT compilation and optimization to accelerate the algorithm. This version of genetic algorithm is used if Numba is installed.
GAlibrate also includes an implementation of the core genetic algorithm that takes advantage of porting some key functions to Julia for JIT compilation and optimization to accelerate the algorithm. This version of genetic algorithm requires Julia and PyJulia; note that the Python-based CLI tool jill is also an option for automating the process of downloading and installing Julia.
GAO runs will display a progress bar that tracks the passage of generations when the tqdm package installed.
PySB is needed to run PySB models, and it is therfore needed if you want to use tools from the `galibrate.pysb`` package.
Tests and coverage analysis use
- pytest (
pytest=7.4.0) - Coverage.py (
coverage=7.2.2)
Running locally from the GAlibrate repo folder:
coverage run -m pytest
then to see coverage report:
coverage report -m
This project is licensed under the MIT License - see the LICENSE file for details
See: CHANGELOG
Principally, GAlibrate defines the GAO (continuous Genetic Algorithm-based Optimizer ) class,
from galibrate import GAOwhich defines an object that can be used setup and run a continuous genetic algorithm-based optimization (i.e., a maximization) of a user-defined fitness function over the search space of a given set of (model) parameters.
The multiprocessing-based parallelism (single node) can be invoked by passing the keyword argument nprocs with a value greater than one when calling the GAO.run function; for example, gao.run(nprocs=2) will use two processes. A full example is provided in this script.
Parallelism is used when evaluating the fitness function across the population (whole population during initialization and half the population during subsequent generations). You can expect the most parallel speedup when the fitness function is expensive to evaluate, such as when evaluating a PySB model. You may also get speedup when the population is very large, depending on how expensive the fitness function is to evaluate. Note however, that if the fitness function is fast to evaluate then the parallel overhead may actually slow down the run.
Additionally, GAlibrate has a pysb sub-package that provides the
galibrate_it module, which defines the GaoIt and GAlibrateIt classes (importable from the galibrate.pysb package level),
from galibrate.pysb import GaoIt, GAlibrateItwhich create objects that abstract away some of the effort to setup and generate GAO instances for PySB models; examples/pysb_dimerization_model provides some
examples for using GaoIt and GAlibrateIt objects. The galibrate_it module can also be called from the command line to generate a template run script for a PySB model,
python -m galibrate.pysb_utils.galibrate_it pysb_model.py output_pathwhich users can then modify to fit their needs.
Additional example scripts that show how to setup and launch Genetic Algorithm runs using GAlibrate can be found under examples.
Email support inquiries to blakeaw1102@gmail.com
See CONTRIBUTING
