.. _modules.changelog: Changelog ========= [1.2.2] - 2026-05-17 ^^^^^^^^^^^^^^^^^^^^ - Added support for ``numpy`` 2 and Python 3.13. Bumped minimum versions of ``scipy``, ``matplotlib``, ``pandas``, ``seaborn``, and ``biopython``. - Added :meth:`~fastdfe.spectrum.Spectrum.scale_theta` and :meth:`~fastdfe.spectrum.Spectrum.get_neutral` methods to the ``Spectrum`` class. - Fixed potential race condition when the same FASTA or GFF files are unzipped simultaneously. - Use approximate equality instead of strict identity when looking up precomputed values for the dominance coefficient ``h``. [1.2.1] - 2026-01-04 ^^^^^^^^^^^^^^^^^^^^ - Don't average likelihood over bootstraps by default as it may mitigate optimization noise but is not statistically sound (see `update_likelihood <../search.html?q=update_likelihood>`_). - Set ``do_bootstrap`` flag to false by default for methods performing nested model comparison. [1.2.0] - 2025-12-24 ^^^^^^^^^^^^^^^^^^^^ - Adding dominance coefficient ``h`` to DFE inference. ``h`` can either be fixed or inferred jointly with the DFE parameters. We can also introduce a relationship between ``h`` and ``S`` (see `h_callback <../search.html?q=h_callback>`_). When estimating ``h``, precomputation over a grid of dominance coefficients takes some time. Implementation was validated with ``SLiM``. - By default, parameters are now fixed to infer a semidominant (``h=0.5``) deleterious DFE without correcting for ancestral misidentification (``eps=0``). - Improved bootstrapping. By default, 2 runs are carried out per bootstrap sample and the most likely result is taken (see `n_bootstrap_retries <../search.html?q=n_bootstrap_retries>`_ which previously controlled the number of retries in case of optimization failure). Bootstrapping is now also carried out by default (`do_bootstrap <../search.html?q=do_bootstrap>`_), and mean and standard deviation across bootstraps are logged. - Initial optimization runs are now recorded in :attr:`~fastdfe.base_inference.BaseInference.runs` dataframe. - Added :class:`~fastdfe.parametrization.DFE` class representing a frozen :class:`~fastdfe.parametrization.Parametrization`. - Expanded documentation on SFS parsing and DFE inference. - Allow to specify how the point estimate is determined when plotting discretized DFE with confidence intervals (see `point_estimate <../search.html?q=point_estimate>`_). - Refactored :class:`~fastdfe.base_inference.InferenceResult`. - Refactored methods returning CIs. For example, removed ``get_cis_params_mle()``, use `get_errors_params_mle() <../search.html?q=get_errors_params_mle>`_ instead. [1.1.13] - 2025-11-22 ^^^^^^^^^^^^^^^^^^^^^ - Fixed bootstrap issue where seeding caused unwanted correlation between the resampled neutral and selected SFS, which could result in smaller confidence intervals. [1.1.12] - 2025-05-26 ^^^^^^^^^^^^^^^^^^^^^ - Made ``cyvcf2`` an optional dependency via ``fastdfe[vcf]`` - Added support for ``Conda`` - Improved x-axis labels for discretized DFE plots - Added support for introducing ancestral misidentification in ``Spectrum`` - Added unnormalized Watterson’s theta property [1.1.11] - 2025-03-30 ^^^^^^^^^^^^^^^^^^^^^ - Support for passing alternative optimizer to ``BaseInference`` and ``JointInference``. [1.1.10] - 2025-03-25 ^^^^^^^^^^^^^^^^^^^^^ - Allow for probabilistic polarization when parsing SFS by looking at ancestral allele probability VCF info tag - Allow the transition/transversion ratio in the ``K2SubstitutionModel`` to be fixed to the value observed in the data - Adjust LRTs to account for parameters near boundaries. The resulting p-values are similar but tend to be somewhat lower - Extend ``ExistingOutgroupFiltration`` so that number of missing outgroups can be specified - Add ``RandomStratification`` and ``ContigFiltration`` classes [1.1.9] - 2025-01-01 ^^^^^^^^^^^^^^^^^^^^ - Add simulation class for simulating SFS data with known DFE [1.1.8] - 2024-08-14 ^^^^^^^^^^^^^^^^^^^^ - Update cyvcf2 dependency to fix broken wheel for Mac ARM (see `issue `_) - Fix minor problem with remote files when disabling file caching [1.1.7] - 2024-05-31 ^^^^^^^^^^^^^^^^^^^^ - Implement serialization of maximum likelihood ancestral annotation to allow for later inspection of results - Improved ancestral allele info tag information for site where annotation was not possible [1.1.6] - 2024-04-20 ^^^^^^^^^^^^^^^^^^^^ - Lazy-load some modules to allow faster initial loading [1.1.5] - 2024-03-12 ^^^^^^^^^^^^^^^^^^^^ - Support for Python 3.12 - Exclude large files from distribution to speed up installation in R [1.1.4] - 2024-03-05 ^^^^^^^^^^^^^^^^^^^^ - Lazy-load some modules to allow faster initial loading [1.1.3] - 2023-12-27 ^^^^^^^^^^^^^^^^^^^^ - Implement probabilistic subsampling for ancestral allele annotation [1.1.2] - 2023-11-27 ^^^^^^^^^^^^^^^^^^^^ - Improved bootstrapping - Set ``allow_divergence`` flag to false by default which as it has the potential to bias the SFS [1.1.1] - 2023-11-21 ^^^^^^^^^^^^^^^^^^^^ - Probabilistic parsing of SFS - Functionality to subsample already existing SFS to lower sample size - Support for number of target sites for ancestral allele annotation - Improved parallelization when bootstrapping - New plotting functionalities - Improved logging [1.1.0] - 2023-10-10 ^^^^^^^^^^^^^^^^^^^^ - Improved parsing utilities - Ancestral allele annotation with outgroups [1.0.0] - 2023-08-12 ^^^^^^^^^^^^^^^^^^^^ - First stable release