Releases: mj-will/nessai
v0.7.0
This is the v0.7.0 release of nessai.
This release supports Python 3.8 to 3.10.
Important: this release changes the flow backend from nflows to glasflow. As a result, the minimum PyTorch version has increased to 1.11.0 and upgrading to this release might also require upgrading PyTorch.
Summary of notable changes
- Explicit support for Python 3.10 (#224)
- Change flow backend from
nflows
toglasflow.nflows
(#228) - Refactor
nessai.nestedsampler
andnessai.flowmodel
into submodules. (#210, #216) - More logging configuration (#229)
- More checkpointing configuration (#225)
- More options for posterior sampling (#213, #214, #233)
- Change
nessai.utils.rescaling.{logit, sigmoid}
to closer matchtorch.{logit, sigmoid}
and changefuzz
toeps
(#218)
Full Changelog: v0.6.0...v0.7.0
v0.6.0
Summary
This release includes various new minor features and bug fixes.
Note: This release removes support for Python 3.6
Main changes
- Remove support for Python 3.6
- Add
nessai.plot.corner_plot
as an alternative toplot_live_points
. - Style for plots is no longer global and can be disabled.
- Add
nessai.config
for configuring global settings innessai
- Add
Model.unstructured_view
for viewing structured arrays (live points) as standard numpy arrays - More options for configuring the logger in
setup_logger
- More examples
- Minor bug fixes.
For more details see the changelog.
v0.5.1
This patch fixes a recently discovered bug related to the initial sampling from the prior.
Details
The bug occurs very infrequently when the prior is difficult to sample from. It therefore shouldn't affect most runs and is very obvious when it does occur. The clearest indication that this bug has occurred is very few posterior samples (<100).
What's Changed
Full Changelog: v0.5.0...v0.5.1
v0.5.0
Summary
This release contains some minor new features, more examples, general improvements to the code base and bug fixes. Most notably:
- Improvements to multiprocessing (now supports user-defined pool)
- Support for vectorised likelihoods
For more details see the full changelog linked below.
Full Changelog: v0.4.0...v0.5.0
v0.4.0
Summary
This release adds a new mode for computing the latent likelihood contour updates various default values to use the new mode (including some defaults for GWFlowProposal
). It also includes a few general improvements and minor features.
Constant Volume Mode
Constant volume mode is an option in FlowProposal
that fixes the volume of the latent contour based on the number of dimensions in the latent space and the value of volume_fraction
. This is the same as fixing the radius of the latent contour.
This significantly improves results for problems where the posterior distribution "rails" against the prior bound and makes results less sensitive to the configuration of the normalising flow. This combined with updated default values means that nessai
is more consistent out-of-the-box with no configuration.
Bilby
Constant volume mode is not supported in the current release of bilby
. A PR with an update to the interface for nessai
is in-progress and should be available in the next release of bilby
.
2022-01-14: Last release of bilby
includes changes to the nessai
interface for this release. (Changelog)
What's Changed
- Add constant volume mode by @mj-will in #125
- Update defaults by @mj-will in #126
- Errors for 1d models by @mj-will in #121
- Add periodic reparameterisation by @mj-will in #127
- Propagate logger by @mj-will in #128
- Use GW reparameterisations by default by @mj-will in #129
- Update GW defaults by @mj-will in #130
Full Changelog: v0.3.3...v0.4.0
v0.3.3
v0.3.2
Minor improvements and bug fixes
This release has a few minor improvements and bug fixes. It also explicitly adds support for python 3.9, which worked previously but was not tested.
See the changelog for details.
v0.3.0 - Testing, testing and more testing
This release contains a large number of changes related to bugs and issues that were discovered when writing more tests for nessai
. It also adds a number of new features and examples.
Note: Runs produced with previous releases are incompatible with this release and cannot be resumed without manual intervention.
See the changelog for further details.
Release 0.2.4
The release includes a number of bug fixes and other minor improvements. For more details see the changelog.