Bugfixes

Fix a bug related to the background scaling on the main observable

Bugfixes

Fixes bugs when evaluating the flux with a background spectrum loaded

Instrument model

We are introducing instrument models for cross calibration between different observatories, and starting a large refactor of the code for the 1.0 release. Also contains several bug fixes.

Full Changelog: v0.2.2...v0.3.0

v0.2.2

What's Changed

• fix plot_ppc with components for fixed parameters by @renecotyfanboy in #219
• patching DiskBB by @renecotyfanboy in #220
• Migrate to uv by @renecotyfanboy in #225

Full Changelog: v0.2.1...v0.2.2

v0.2.1

Upgrades

• Better documentation on how to mock data without explicit observation
• fakeit_for_multiple_parameters has been optimized to explicitly prevent the replication of response matrix, easing folding for multiple parameters at once without blowing up memory
• prior_predictive_coverage should use multiple cores now
• Internal cleaning : we do not depend explicitly on the Heacit dataset but rather download what is needed on the fly

Bug Fixes

• Fix tbpcf and FDcut models that were not fully ported after the internal rework of jaxspec
• Fix Agauss, Zagauss, Zgauss which were in a similar situation

Full Changelog: v0.2.0...v0.2.1

v0.2.0

This new version brings a brand-new backend for the jaxspec model building tool. We are switching from haiku (which is now in maintenance mode) to nnx to handle the parameters in JAX. It brings some new breaking changes, but will help a lot for future maintenance.

Upgrades

• Background model scaling has been fixed, there was an issue because the scaling factor was applied after folding while it should be before.
• plot_ppc can now display multiple components in the model, see here
• Codebase and documentation were cleaned a bit

Breaking changes

• Model parameters are now all lower case and underscore use is forbidden for now. If you used components such as tbabs, be sure to change N_H to nh
• Custom model components should be built using nnx. It changes a bit when compared to haiku, and has been updated accordingly in the documentation

What's Changed

• Update pytest-cov requirement from >=4.1,<6.0 to >=4.1,<7.0 by @dependabot in #202
• Factor ARF from RSP by @lmauviard in #205
• switching to nnx by @renecotyfanboy in #203
• Update mendeleev requirement from >=0.15,<0.19 to >=0.15,<0.20 by @dependabot in #207
• Update mkdocstrings requirement from >=0.24,<0.27 to >=0.24,<0.28 by @dependabot in #206
• updating dependencies by @renecotyfanboy in #208
• Update numpyro requirement from ^0.15.3 to >=0.15.3,<0.17.0 by @dependabot in #211
• Update ruff requirement from >=0.2.1,<0.8.0 to >=0.2.1,<0.9.0 by @dependabot in #210
• update documentation by @renecotyfanboy in #209

New Contributors

• @lmauviard made their first contribution in #205

Full Changelog: v0.1.4...v0.2.0

v0.1.4

What's new ?

• Fixed a bug where subtracting background had no effect
• Add the possibility to use a spectral model as the background
• Luminosity can be computed using either distance or redshift
• Prior predicting coverage is clearer

Full Changelog: v0.1.3...v0.1.4

v0.1.3

Just fix a silent issue in the previous release

v0.1.2

• Prior predictive coverage using the BayesianFitter
• Internal fixings
• Pretty representation for SpectralModel

v0.1.1

This is our first release after the article acceptance.

• All the numpyro's MCMC are now gathered within our MCMCFitter class, where the user can switch using the sampler parameter
• Priors can now be passed as a flat dictionary instead of the nested dictionary from haiku. We want it to be the main way to pass prior in the future releases.
• Changes in the inner workings of the fitter for future features.