- Added tested source distributions, pure-Python wheels, and SHA256 checksums to GitHub Releases.
- Added Git-archive version metadata for
setuptools-scmso tagged archives can recover their exact version without a.gitdirectory. - Added CI validation of checkout and Git-archive builds, isolated source-distribution rebuilds, clean wheel installations, scripts, and packaged scientific resources.
- Updated release automation and installation documentation to use the validated GitHub Release artifacts as the canonical distributions.
- Added support for correlation functions binned in (r, mu) and Legendre multipole compression, including covariance and distortion matrices and metal correlations.
- Added direct input of QSO auto-correlation multipoles and their covariance for joint Ly-alpha auto, Ly-alpha-QSO cross, and QSO auto analyses, with consistent separation cuts and multipole selection in the global covariance.
- Added catalog-based redshift weights and bias evolution for QSO multipoles and cross-correlations.
- Added support for externally computed two-dimensional P(k, mu) models, including metal contributions, with example notebooks for custom models and configurations.
- Added the aiso/aap Alcock-Paczynski parametrization used in DESI galaxy analyses.
- Added optional Hartlap corrections for individual and global covariances, with the Percival correction applied to saved parameter errors and covariance for global fits using the Hartlap correction.
- Extended FITS output and fit postprocessing for multipoles, fit p-values, and covariance corrections; added multipole plotting and improved four-wedge plots.
- Fixed fiducial-model masking and Monte Carlo mock generation with global covariances, and improved coordinate-binning checks.
- Added a model-binning option to the configuration builder and updated small-scale marginalization parameter handling.
- Added tests and example configurations for QSO multipoles and catalog redshift weights, updated the README, and required SciPy >= 1.12.0.
- Migrated package setup from setup.py to modern pyproject.toml standard.
- Replaced flake8/isort with Ruff and implemented pre-commit hooks (need to be activated).
- Updated CI/CD pipeline to use modern GitHub Actions with dependency caching.
- Refactored documentation to use sphinx-book-theme and MyST-NB for Jupyter notebook rendering.
- Replaced hardcoded version strings with dynamic setuptools_scm versioning.
- Version used for DESI DR2 full-shape KP
- Minor updates to metal parameters
- Catch PackageNotFoundError when initializing Vega
- Some protections for log_lik and marg_coeff behaviour
- Added printing of vega version when initializing VegaInterface
- Some protections for log_lik and marg_coeff behaviour
- New modelling options for binning
- New modelling options for HCDs
- Minor bug fix in ConfigBuilder
- Added UVB shotnoise and HeII reionization models
- Updated some flag names for consistency
- Added support for the redshift bin analysis
- Added new options for small-scale marginalization:
- Option to fit the marginalized bins
- Option to marginalize over data bins instead of model bins
- Option to fit marginalized parameters at every step of the likelihood instead of marginalizing in the covariance matrix.
- Updates and improvements to the small-scale marginaliztion functionality
- Improvements to the sampler performance
- Add SSM parameters as derived parameters in PolyChord
- Unblind DR2 alpha_smooth parameter
- Add support for alpha_full analyses while phi is still fitted separately
- Some minor fixes to the low memory mode for sampler runs
- Add analytic marginalization over small-scale marginalization parameters
- Fix bug with QSO-metal computation of metal matrices
- Optimize metal computation by caching more correlations
- Added small scale marginalization for distorted analyses
- Add option for k^n smoothing to match EFT behavior
- Minor updates to the ConfigBuilder
- Minor bug fixes
- Unblinded DR2 CIV BAO analysis
- Added Arinyo q2 parameter
- Added explicit call limit for iminuit
- Minor bug fixes
- Add support for shell plots in the plotting module
- Automatic shell plots after fits when using run_vega.py
- Add option to fit independent full-shape smoothing parameters for each tracer
- Add MC functionality to the ConfigBuilder
- Some minor bug fixes related to MC mocks
- Overhaul of the Monte Carlo mocks functionality, now supports global covariances
- Minor improvements to blinding functionality, and added blinding for DR2 full-shape
- Update flat priors to have more realistic ranges
- Added options to compute shell compression as a function of mu, mu^2, and theta
- Fixed a bug in the plotting of the Fisher information as a function of rp and rt
- Unblind Y3 BAO analysis
- Some small improvements to FitResults and full-shape blinding
- Some updates to MC mocks. Now it is possible to read a previous fit and use it to generate mocks.
- Updated metal formatrix computation allowing for rt dependecy.
Previous configuration can still be used, as the new one is quite slow to initialize.
- Fixed metal caching bug. Only metal autos are cached now.
- Updated blinding for Y1 and Y3 full-shape analyses
- Added blinding for DESI Y3
- BAO blinding still done through a blinding template
- AP blinding done at the parameter level in Vega
- Added two AP blinding files for DESI Y1 and DESI Y3
- Added AP blinding for DESI DR1 (can only be run at NERSC)
- Updated some variable names to reflect the DESI BAO paper
- Added option to run Monte Carlo mocks from the global covariance
- This is the version used to run the DESI Y1 BAO analysis
- New template: Planck18/DESI-2024_z_2.33.fits
- The growth_rate parameter is now automatically read from the template file by default. This means the value passed will be ignored from now on. There is a way to turn this off by setting use_template_growth_rate=False option in [control].
- Added option to use bias_metal (instead of bias_eta_metal) in ConfigBuilder
- Defaults values updated: beta_CIV=0.5, alpha_CIV=0
- We now recommand sampling bias_metal instead of bias_eta_metal
- Smoothing parameters are now fixed to input value for metal correlations. This fixes the bug that made fits fail when varying smoothing parameters with fast_metals turned on. (only matters for mocks)
- Growth rate is now fixed for metal correlations by default (only matters for full-shape analyses).
- Fixed a bug where the DESI instrumental systematics model was being added twice. This only impacts the value of the fitted amplitude for that model.
- Monte Carlo mocks now support the use of the full covariance matrix.
- Added option to turn off extrapolation when doing the FHT to go from xi to pk in the make_template script. We recommend turning this off when computing templates for non-Planck cosmologies.
- Added example config files for the DESI Y1 analysis under examples/DESI_data_setup for runs on data, and under examples/DESI_mock_setup for runs on mocks.
- DESI Y1 is now unblinded. Vega supports both DA_BLIND and DA column names for desi_y1 blinding
- Distortion matrix blinding status is now ignored
- Added new template at the effective redshift of the DESI Y1 correlations
- Update method used to compute the effective redshift
- Added new metal configuration to the Config Builder
- Added new low memory mode for sampler runs
- Added option to use the full covariance matrix, including all cross-covariances
- New metal modelling:
- Metal matrices are now computed on the fly in vega
- To compute metal matrices vega now requires weights
(delta-attributes file for forests and catalogs for discrete objects) - Backwards compatibility with old metal matrices is maintained
New Monte-Carlo mock functionality, including a new MPI parallelized script
Overhaul of coordinate handling in vega
- New more accurate model for desi instrumental systematics
- Added option to skip all metal auto-correlations (turned off by default)
- Small fixes for non-standard model binning
- Metal smoothing for mocks
- Fixed interpolation bounds for fvoigt model
- Update HCD modelling config and defaults
- To use the Fvoigt model, use keyword "fvoigt" instead of the old "mask"
- "L0_hcd" parameter only applies to Rogers model from now on
- Fvoigt model now has its own "L0_fvoigt" parameter, by default set to 1
- Add parameter sensitivity and information calculations and plots.
- Minor fix for vega installations from tarballs instead of git
- Minor updates to maintain future compatibility with numba
- Fixed some minor issues from v0.4.0
- First version used for the DESI Y1 analysis.
- New plotting module
- New metal computation and smart chaching
- First version used in DESI, and for most of the early analyses. Includes blinding.
- First version of Vega