Release notes v0.2

Release notes v0.2#

Specific upgrades for the latest release can be found here.

New features since the last release#

  • Multi-parameter-of-interest support: poi_test now accepts a Dict[Union[int, str], float] in addition to a plain float. When a dictionary is supplied, every entry fixes the corresponding parameter (identified by index or by name as recorded in ModelConfig.parameter_names) during the fit, while the remaining parameters are optimised freely. A plain float retains exactly the previous single-POI behaviour. The new type alias PoiTest is exported from spey.interface.statistical_model for type-annotation convenience. Affected interfaces: StatisticalModel.likelihood, StatisticalModel.asimov_likelihood, StatisticalModel.fixed_poi_sampler, StatisticalModel.sigma_mu_from_hessian, HypothesisTestingBase.chi2, HypothesisTestingBase.sigma_mu, HypothesisTestingBase.exclusion_confidence_level, and the corresponding methods in UnCorrStatisticsCombiner.

  • ModelConfig gains two new helpers: resolve_poi_indices (converts a poi_test value to a {param_index: value} mapping) and fixed_poi_bounds_multi (generalises fixed_poi_bounds to multiple simultaneously fixed parameters).

  • optimizer.fit extended to accept fixed_poi_value as a Dict[int, float], fixing an arbitrary set of parameters in a single optimisation call.

  • Ability to compute two-sided tests added. (#45)

  • Add plug-in registry that does not require creating an entire package. (#60)

  • Iminuit is available to be used for optimisation as alternative to scipy. (#59)

Improvements#

  • Autograd has been upgraded to v1.7 which supports numpy 2.0 (#50)

  • Enabled user control to allow negative signal yields in \(\chi^2\)-test. (#50)

  • Implement python version compatibility check for new releases. (#52)

  • Added a helper function to merge correlated bins. (#53)

  • Some errors are reduced to warnings in chi2 test to reduce the verbosity. (#54)

  • Enable compatibility for Python 3.13, and the deprecation of pkg_resources has been addressed. (#56)

  • Multivariate Normal distribution now supports callable covariance matrix function. (#59)

  • MultivariateNormal backend now accepts signal_yields as a callable with signature (extra_pars: np.ndarray) -> np.ndarray in addition to a plain array. A companion keyword argument n_signal_parameters (default 0) declares how many free parameters beyond :math:\mu the callable requires; they are appended to the optimiser parameter vector as signal_par_0, signal_par_1, … and are included in ModelConfig with (None, None) bounds. The plain-array path is unchanged.

  • maximize_likelihood and maximize_asimov_likelihood gain a new optional keyword argument poi_indices (Optional[List[Union[int, str]]], default None). When None the methods return the usual (muhat, nll) tuple. When a list of parameter indices or names is provided they return (Dict[Union[int, str], float], nll) with the fitted values for each requested parameter. Supported on StatisticalModel and UnCorrStatisticsCombiner.

  • All default-PDF backends (DefaultPDFBase, UncorrelatedBackground, CorrelatedBackground, ThirdMomentExpansion, EffectiveSigma) now accept a signal_parameter_bounds keyword argument (Optional[List[Tuple[Optional[float], Optional[float]]]], default None). When provided, each entry supplies the (lower, upper) optimiser bound for the corresponding extra signal parameter (indexed by position in the parameter vector immediately after :math:\mu); a None element leaves that side unbounded. When the argument is None (the previous behaviour), every extra signal parameter continues to receive (None, None) bounds. The list must have exactly n_signal_parameters entries. These bounds are stored in ModelConfig.suggested_bounds and propagated to the optimiser automatically.

  • For pdf_type="multivariategauss" with a non-callable cov, the old code was: self._pdf = lambda pars: MultivariateNormal(mean=loc(pars), cov=cov) which re-instantiated MultivariateNormal (and re-ran np.linalg.inv + np.linalg.slogdet, both O(n³)) on every likelihood call. The fix creates a single MultivariateNormal once at construction time and only updates _mv.mean = loc(pars) on each call — the cached _inv_cov /_logdet_cov closures are reused unchanged. When cov is callable, the original per-call re-instantiation is preserved.

  • Minor inefficiencies fixed Normal.log_prob: value[self.domain] was sliced twice; now sliced once into x. MultivariateNormal.log_prob: value[self.domain] was sliced four times; now sliced once into x. Removed unused from copy import deepcopy import.

Bug Fixes#

  • Control mechanism added in case of infinite determinant in covariance matrix. (#47)

  • Error during signal uncertainty insertion due to valid domain has been fixed. (#51)

  • A python version depended bug fixed in about function. This bug only effected python v3.9 and below. (#61)

Contributors#

This release contains contributions from (in alphabetical order):