Changelog

All notable changes to this project will be documented in this file.

[Template]

Added

• Feature X

• Feature Y

Changed

• Refactored module A to improve performance

Fixed

• Bug in module B that caused a crash

[Unreleased]

Added

Changed

Fixed

[0.10.0] - 2024-11-22

Added

• Added the AffineTransformation class to perform affine transformations on the data using jax

• Added the method function forward_vectorized to the (Base)Model class and a specialized jax version in the JaxModel class.

Changed

• Refactored the module eulerpi.core.data_transformation into the subpackage eulerpi.core.data_transformations. The data transformation classes can be import from this subpackage directly

• Renamed eulerpi.core.data_transformation.DataNormalizer into eulerpi.core.data_transformations.DataNormalization

• Removed the class/factory functions from_data and from_transformation from the DataTransformation subclasses. Use the constructors instead.

• The inference function no longer has the parameter custom_data_transformation. You can pass an instance of the provided DataTransformation subclasses or pass a custom subclass.

• The inference function no longer has the parameter n_components_pca. Instantiate a PCATransformation with the data and the number of components to keep. Pass it as as argument for the parameter data_transformation in the inference function.

• Renamed the abstract base Model class to BaseModel. The usage of Model is still supported, but deprecated, and possibly removed in a future release.

• Splitted the file model.py into multiple modules, one for each class, located in the subpackage eulerpi.core.models. The imports change from eulerpi.core.model to eulerpi.core.models.

• Removed the function is_artificial from the (Base)Model class. Use isinstance(custom_model, ArtificialModelInterface) or issubclass(CustomModel, ArtificialModelInterface) instead.

[0.9.1] - 2024-09-08

Changed

• Raised required amici version to 0.26.1

• Ignoring sampler warnings caused by parameters outside the parameter limits

• Limited the parameter range and increased the solver accuracy of the Corona model to avoid warnings

Fixed

• Seed numpy random number generator for reproducibility of pytest runs

[0.9.0] - 2024-05-31

Changed

• Raised the required python version to 3.10

• Removed upper limit on the version number of dependencies

• Updated amici to newest version to fix compilation with Intel MKL

[0.8.1] - 2024-05-31

Added

• Added PyQt6 as dev dependency to allow developers to view plots in the development environment

Changed

• Removed the redundant function evaluate_sample from sampling.py

Fixed

• Switched from importlib.resources.path to importlib.resources.files due to deprecation

• Removed argument name in call to files(...) that is changing from package to anchor

• Removed unused kwargs passed to the plotting function contour that generated a warning in test_heat_model()

• Bug that changed the model attribute central_param during inference

[0.8.0] - 2024-03-21

Added

• Basic plotting function for sample results

• Users can now check the models they want to use with a basic functionality check and a quick inference check on artificial data.

• Added code usage examples to documentation.

Fixed

• Fixed deprecated jax.config import

[0.7.0] - 2023-11-03

Breaking changes: The stock model is removed from the available example models! You can copy the model code and the data files from a previous version and import it as custom model, if you need it.

Changed

• Removed the stock model, because it does not fulfill the requirements for the EPI method

• Minor dependency updates

[0.6.1] - 2023-11-02

Added

• Added link to changelog on pypi

[0.6.0] - 2023-11-02

Breaking changes: SBML support is now only available when install eulerpi with the “sbml” extra!

Added

Changed

• Updated the dependencies to the latest possible versions

• Made the sbml model support optional using the extra sbml

Fixed

• Fixed poetry.lock containing faulty version 1.6.8 of debugpy

• Fixed github actions failing due to missing sudo apt update

[0.5.0] - 2023-07-20

Added

• Added support for for sbml models to select parameters and species used in the inference

• Added support for evaluating sbml models at multiple time points

Changed

• Switched from using parameter names to using parameter ids in the sbml model

• Change spatial discretization of heat model example to proper second order central differences

• Change the initial walker position generation of the emcee sampler

[0.4.0] - 2023-06-24

Added

• Added a new example model, a 2d heat conduction equation

• Added a function to model to specify more complex parameter domains

Changed

• Removed old functions from corona model

• Temporarily removed stock and sbml models from test examples

Fixed

• Updated tutorial to be consistent with current state of the project

• Fixed bug in dense grid generation causing parameter limits to be applied inconsistently when using slices

[0.3.1] - 2023-05-02

Fixed

• Bug in result manager causing burn-in and thinning to be performed on the wrong samples.

• Bug in result manager that caused density evals to be saved as data samples for non-full slices.

[0.3.0] - 2023-04-27

Added

• Added data normalization to improve numerics and performance in the inference method.

• Added possiblity to use no PCA transformation or completely custom transformation for inference.

• More tests to validate EPI

Fixed

• Fixed grid results beeing unusable due to numerical issues

• Fixed sbml models ignoring parameter_names argument

[0.2.1] - 2023-04-21

Fixed

• Fix sparse grid returning meaningless data

• Fix menten data shape

• Fix SBMLModel data_dim: Return states count instead of observables count

Added

• Validating the shapes of the inference results

[0.2.0] - 2023-04-20

Changed

• Now using default emcee move policy during MCMC sampling to improve convergence speed

• Changed argument calc_walker_acceptance_bool to get_walker_acceptance

• Changed meaning of argument num_burn_in_samples to refer to the number of samples burned per chain, not in total

Fixed

• Fixed SwigPyObject pickling

• Fixed dead lock for larger data sets and models

Added

• ResultManager is saving meta data for inference

• ResultManager uses saved meta data to load inference results

[0.1.5] - 2023-04-06

Changed

• Renamed epipy to eulerpi

[0.1.4] - 2023-04-06

Changed

• Updated installation instructions

[0.1.3] - 2023-04-05

Changed

• Updated dependencies

[0.1.2] - 2023-04-05

Fixed

• removed __pycache__, build and .so from pypi

[0.1.1] - 2023-04-05

Fixed

• fixed epi logo not showing on pypi

[0.1.0] - 2023-04-05

Added

• Initial release