Cobaya

General

- Updated UGE sample job submission template (for cobaya-job-run and cobaya-grid-run)
- Clarify log feedback when using oversample_thin
- Fixed 345, 346, 347, 348

Grid scripts
- Support for running grids of models, including grid getdist, PDF tables, importance sampling, minimization (almost all features of CosmoMC grid now available in Cobaya). See the new doc pages.

Minimization
- Support for iminuit minimizer and getting best-fits for all mpi runs (332, thanks ggalloni)
- Support for minimzation with an importance-sampled input yaml config

General

- Fixed a packaging bug after migration to `pyproject.toml`

Cosmology

CLASS

- Min version update to 3.2.1 (solves 305; thanks to the CLASS developers)

General

- Python 3.12 support (removed all dependence on distutils)
- Improved `.products()` method for samplers (MCMC and PolyChord) and post-processing: samples can now retrieved simultaneously for all MPI processes, and converted to GetDist. Also added `.samples()` methods to retrieve just the samples.

Collections

- Created a general `load_samples` function to load Cobaya results natively or as GetDist MCSamples.
- Collections are now aware of whether they are part of a parallel batch, and warn if trying to reweight/detemper individually (fixes 321).
- Fixed a bug with overzealous checks when loading samples (306, thanks mishakb for reporting).

MCMC

- Fixed a bug with mpi runs partly stalling when run with many chains (308, thanks vivianmiranda lukashergt for reporting and testing).
- When `oversample_thin` is used, avg thinned weights now reported instead of acceptance rate (310, thanks vivianmiranda for reporting)

Cosmology

BAO

- Added 1-d grid LSS likelihood and BAO-only ELG and QSO (PR 266; thanks msyriac)

CLASS

- Updated manual installation instructions and fixed some dependencies.
- Made more derived parameters available, and documented how to access even more.
- Fixed 292: wrong normalization for the Cl cross-spectra (thanks carlosggarcia)

General

- Class instance methods can now be used as external likelihoods.
- Fix _prior_tries_warning bug
- Fix over-stringent temperature test reading in chains

PolyChord

- `products` method revamped; can produce GetDist chains directly.

Cosmology

- updated CAMB min version to 1.5, fixing bug with Cobaya sampling
- cobaya-install cosmo now installs set of Planck NPIPE (PR4) python likelihoods
- added planck_2018_lowl.EE_sroll2 low-E Planck likelihood
- added startup warning if initial points are very over-dispersed compared to the proposal covariance
- Requesting CAMBdata from camb now a copy for exact initial power spectrum/non-linear model
- CAMB now supports using sigma8 as an input parameter (thanks tilmantroester)

- Updates for Pandas 2 compatibility
- Fixed bug in MCMC oversampling and simplified proposal code (288) (thanks JiangJQ2000)

General

- Minimum Python version updated to 3.8
- `Prior.bounds()` can now return bounds at particular confidence levels when passed `confidence<1`.
- `SampleCollection` slicing now allows for advanced pandas slicing, e.g. `samples[samples["param"] > value]`.
- Fixed bug when setting reference pdf in MPI runs (thanks schoeneberg!)
- Components in yaml files referring to external Python modules can now give `package_install` settings to specify whether installed from pip, github or URL when cobaya-install is run.
- Fix for `post` when likelihoods return different number of derived parameters (285) (thanks zhaoruiyang98)

MCMC

- Added tempered sampling.
- `products` method revamped; can produce GetDist chains directly.

Cosmology

- Replaced default planck_2018_lowl.EE and planck_2018_low.TT with native versions, and using GitHub-hosted clik version.
- Updated planck likelihoods to all load calibration parameter from same yaml
- Removed clik version of planck 2018 CamSpec, defaults to native (avoids inconsistent calibration parameter naming)
- GUI inclues latest NPIPE fully Python likelihood configuration