Skip to content

Latest commit

 

History

History
105 lines (87 loc) · 12.1 KB

CHANGELOG.md

File metadata and controls

105 lines (87 loc) · 12.1 KB
Raw

Changelog

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

Unreleased

Added

  • #1500 Added a CensoredGaussianLogLikelihood class that calculates the censored Gaussian log-likelihood.
  • #1505 Added notes to ErrorMeasure and LogPDF to say parameters must be real and continuous.
  • #1499 Added a log-uniform prior class.

Changed

  • #1503 Stopped showing time units in controller logs, because the units change depending on the output type (see #1467).

Deprecated

Removed

Fixed

  • #1517 Fixed a major bug in the covariance matrix update for xNES.
  • #1505 Fixed issues with toy problems that accept invalid inputs.

[0.5.0] - 2023-07-27

Added

  • #1484 Added a GaussianIntegratedLogUniformLogLikelihood class that calculates the log-likelihood with its Gaussian noise integrated out with an uninformative prior in log-space.
  • #1466 Added a TransformedRectangularBoundaries class that preserves the RectangularBoundaries methods after transformation.
  • #1462 The OptimisationController now has a stopping criterion max_evaluations.
  • #1460 #1468 Added the Adam local optimiser.
  • #1459 #1465 Added the iRprop- local optimiser.
  • #1456 Added an optional translation to ScalingTransform and added a UnitCubeTransformation class.
  • #1432 Added 2 new stochastic models: production and degradation model, Schlogl's system of chemical reactions. Moved the stochastic logistic model into pints.stochastic to take advantage of the MarkovJumpModel.
  • #1420 The Optimiser class now distinguishes between a best-visited point (x_best, with score f_best) and a best-guessed point (x_guessed, with approximate score f_guessed). For most optimisers, the two values are equivalent. The OptimisationController still tracks x_best and f_best by default, but this can be modified using the methods set_f_guessed_tracking and f_guessed_tracking.
  • #1417 Added a module toy.stochastic for stochastic models. In particular, toy.stochastic.MarkovJumpModel implements Gillespie's algorithm for easier future implementation of stochastic models.
  • #1413 Added classes pints.ABCController and pints.ABCSampler for Approximate Bayesian computation (ABC) samplers. Added pints.RejectionABC which implements a simple rejection ABC sampling algorithm.
  • #1378 Added a class pints.LogNormalLogLikelihood.

Changed

  • #1485 PINTS is no longer tested on Ubuntu 18.04 LTS, but on 20.04 LTS and 22.04 LTS.
  • #1479 PINTS is no longer tested on Python 3.6. Testing for Python 3.11 has been added.
  • #1479 The asyncio.coroutine decorators have been removed from all of NUTS's coroutines in order to be compatible with Python 3.11.
  • #1466 Transformation.convert_boundaries will now return a TransformedRectangularBoundaries object if the transformation is element-wise and the given boundaries extend RectangularBoundaries.
  • #1458 The GradientDescent optimiser now sets its default learning rate as min(sigma0) (it can still be changed afterwards with set_learning_rate()).
  • #1445 Allowed multiple LogPDFs to be supplied to the MCMCController (one for each MCMC chain), and added an evaluator which evaluates each position on a separate callable.
  • #1439, #1433 PINTS is no longer tested on Python 3.5. Testing for Python 3.10 has been added.
  • #1435 The optional Stan interface now uses (and requires) pystan 3 or newer. The update_data method has been remove (model compilation is now cached so that there is no performance benefit to using this method).
  • #1424 Fixed a bug in PSO that caused it to use more particles than advertised.
  • #1424 xNES, SNES, PSO, and BareCMAES no longer use a TriangleWaveTransform to handle rectangular boundaries (this was found to lead to optimisers diverging in some cases).

Removed

  • #1424 Removed the TriangleWaveTransform class previously used in some optimisers.

Fixed

  • #1497 Fixed deprecation warning of np.product globally in pints.
  • #1457 Fixed typo in deprecation warning for UnknownNoiseLikelihood.
  • #1455 The s and inv_s properties of ScalingTransformation have been replaced with private properties _s and _inv_s.
  • #1450 Made TransformedBoundaries consistent with Boundaries by removing range() and adding sample().
  • #1449 Fixed a bug in MarkovJumpModel.interpolate_mol_counts.
  • #1399 Fixed a bug in DramACMC, and fixed the number of proposal kernels to 2.

[0.4.0] - 2021-12-07

Added

  • #1409 The OptimisationController now accepts a callback function that will be called at every iteration; this can be used for easier customisation or visualisation of the optimiser trajectory.
  • #1383 Added a method toy.TwistedGaussianDistribution.untwist that turns samples from this distribution into samples from a multivariate Gaussian.
  • #1322 Added a method sample_initial_points that allows users to generate random points with finite metrics (either log-probabilities or error measures) to use as starting points for sampling or optimisation.
  • #1243 Added testing for Python 3.9.
  • #1213, #1216 Added the truncated Gaussian distribution as a log prior, TruncatedGaussianLogPrior.
  • #1212 Added the PooledLogPDF class to allow for pooling parameters across log-pdfs.
  • #1204 This CHANGELOG file to show the changes introduced in each release.
  • #1190 A new ConstantAndMultiplicativeGaussianLogLikelihood was added.
  • #1183 Three new methods were added for diagnosing autocorrelated or time-varying noise: plot_residuals_binned_autocorrelation, plot_residuals_binned_std, and plot_residuals_distance.
  • #1175 Added notebooks showing how to interface with the statsmodels Python package which allows fitting ARIMAX and state space models in PINTS.
  • #1165 A new Transformation abstract class was added, along with ComposedTransformation, IdentityTransformation, LogitTransformation, LogTransformation, RectangularBoundariesTransformation, ScalingTransformation subclasses to achieve more effective and efficient optimisation and sampling.
  • #1165 A new optional argument transform was added to both OptimisationController and MCMCController to transform parameters during optimisation and sampling.
  • #1112 A new NoUTurnMCMC sampler (NUTS) was added, along with a DualAveragingAdaption class to adaptively tune related Hamiltonian Monte Carlo methods.
  • #1025 Added a stochastic logistic growth problem for use with ABC.

Changed

  • #1420 The OptimisationController now logs a best and a current score.
  • #1375 Changed all arguments called transform to transformation for consistency.
  • #1365 Dropped support for Python 2.7. PINTS now requires Python 3.5 or higher.
  • #1360 The ParallelEvaluator will now set a different (pre-determined) random seed for each task, ensuring tasks can use randomness, but results can be reproduced from run to run.
  • #1357 Parallel evaluations using multiprocessing now restrict the number of threads used by Numpy and others to 1 (by default).
  • #1355 When called with parallel=True the method pints.evaluate() will now limit the number of workers it uses to the number of tasks it needs to process.
  • #1250 The returned values from SingleChainMCMC.tell() and MultiChainMCMC.tell() have been extended from current position x to x, fx, accepted, where fx is the current log likelihood and accepted is a bool indicating whether tell performed an acceptance step in this call.
  • #1195 The installation instructions have been updated to reflect that PINTS in now pip-installable.
  • #1191 Warnings are now emitted using warnings.warn rather than logging.getLogger(..).warning. This makes them show up like other warnings, and allows them to be suppressed with filterwarnings.
  • #1112 The pints.Logger can now deal with None being logged in place of a proper value.

Deprecated

  • #1420 The methods pints.Optimisation.xbest() and fbest() are deprecated in favour of x_best() and f_best().
  • #1201 The method pints.rhat_all_params was accidentally removed in 0.3.0, but is now back in deprecated form.

Removed

  • #1250 The methods SingleChainMCMC.current_log_pdf() and MultiChainMCMC.current_log_pdf() have been removed.

Fixed

  • #1350 Fixed bugs in the Relativistic MCMC sampler.
  • #1264 Fixed a bug relating to how NUTS handles nans when values outside the range of the priors are proposed.
  • #1257 Fixed a bug in GaussianLogPrior, which meant the distribution could be instantiated with a non-positive standard deviation.
  • #1246 Fixed a long-standing bug in PopulationMCMC, which caused it to sample incorrectly.
  • #1196 The output of the method pints.HalfCauchyLogPrior.sample had the wrong shape.

[0.3.0] - 2020-08-08

  • This is the first pip installable release. The changelog documents all changes since this release.