v5.2.1

Octofitter v5.2.1

Diff since v5.2.0

Bug Fixes

• Fixed a major bug in ObsPriorAstromONeil2019, an available prior transformation based on O'Neil et al. (2019).

v5.2.0

Octofitter v5.2.0

Diff since v5.1.0

Improved plotting

Breaking changes

• Plots: remove planet_rv argument to octoplot. It was never documented or well supported.

New Features and Improvements

• Plots: add dark band for GP std per draw in octoplot
• Support plotting GPs in octoplot
• Plots: Improve rv octo plot to show multiple panels
• Plots: improve rvpostplot performance on large chains
• Plots: visual improvements to rvpostplot
• Improve Dynesty.jl integration
• Plots: remove grid lines in rvpostplot for cleaner look
• Plots: Centre RV phase-folded plot at zero crossing

v5.1.0

Octofitter v5.1.0

Diff since v5.0.0

Release notes:

New features:

• Support for rvpostplot for plotting multi-planet fits
• Optional “perspective” line for Absolute RV fits

Bug fixes:

• Plots: Use mas instead of arcsec for tight orbits, as intended
• Plots: Allow rvpostplot to work with only one datapoint
• Cross validation: Fix RV epoch subsetting
• Whereistheplanet loader: Forward catalog to search function
• Improved warning and error messages

Closed issues:

• Global Main._obs error in octofitterpy (#49)

v5.0.0

Octofitter v5.0.0

Diff since v4.0.0

The RV Release!

Breaking changes:

• New RV API
  • The RV likelihood objects much now be created separately per-instrument (no more inst_idx).
  • The jitter and/or offset parameters can be given different names per likelihood object, or shared
  • Can now use MarginalizedStarAbsoluteRVLikelihood to analytically marginalize over an instrument RV zero-point
  • gaussian processes: GP now fit per instrument instead of globally, matching behaviour of RadVel
  • Can now fit an RV trend with any arbitrary function, per instrument or globally

Other changes

• Improved performance on large datasets
  • Kepler solver now multithreaded
  • Threads are used automatically if Julia started with --threads=auto
  • Threads are used with the HMC sampler (octofit)
  • Threads are used with Pigeons, if using MPI and more than one thread is available per-worker
• Improved automatic sampler initialization substantially
• Improved RV plot, which now shows a histogram of residuals
• More control over HGCA likelihood. Can fit an instantaneous PMA, or average over arbitrary # of epochs
• Can now plot orbits in physical AU units, in addition to astrometric units (e.g arc seconds)
• Many small improvements to plots
• Pigeons integration now defaults to SliceSampler
• New cross validation functionality for testing impacts of each instrument, or each data point:
  • given one model with data from N instruments, automatically generate N new models that each drop one instrument's data

Merged pull requests:

• Use multi-threading in likelihood (#44) (@sefffal)

Closed issues:

• Not able to retrieve planet properties (#45)

v4.0.0

Octofitter v4.0.0

Diff since v3.1.0

Breaking changes

• RV zero points and jitters must now be specified as array-valued distributions. E.g. rv0 ~ MvNormal([100,100]) instead of rv0_1 ~ Normal(100); rv0_2 ~Normal(100).

New Features

• More robust, and more efficient initialization of chains
• KDE priors
• Multi dimensional / correlated priors, e.g. can sample from a spectrum using a multivariate normal list.
• Much improved Hipparcos IAD modelling
• RV Modelling can now account for perspective acceleration
• Precision Improvements: more accurate treatment of time throughout code, now suitable for very short period orbits
• Improved plots
• Support for closure phase / kernel phase modelling of GRAVITY raw data.
• Support for nested sample with Dynesty.jl (ONLY for sake of performance comparison to Pigeons, not recommended)
• Added ability to sample from x vs y positions, for single epoch observations (e.g. detection maps)
• Added ability to use Octofitter to extract classical photometry and astrometry from images
• Plotting style improvements
• Probabilistic programming language: can now interpolate global variables into the model definition using $
• Additional error checking and robustness
• Warnings generated for type-instability in models that could cause poor performance

Merged pull requests:

• CompatHelper: add new compat entry for HTTP at version 1, (keep existing compat) (#38) (@github-actions[bot])
• CompatHelper: add new compat entry for Distances at version 0.10, (keep existing compat) (#39) (@github-actions[bot])
• CompatHelper: add new compat entry for CSV at version 0.10, (keep existing compat) (#40) (@github-actions[bot])

Closed issues:

• Jupiter mass definition should follow IAU standard (#34)
• Issue using RV data with octofitterpy (#41)

v3.1.0

Octofitter v3.1.0

Diff since v3.0.2

• Fixes a regression in Pigeons by upper-bounding the version compat
• Plotting improvements (style tweaks and performance improvements)
• Use BJD instead of JD for both HARPS RV data sources
• Export KDEDist constructor for kernel density estimates
• Improve KDEDist constructor with optional upper and lower bounds
• 2-4x improvement in HGCA modelling performance
• Improvement to pathfinder-based sampler initialization (both HMC and Pigeons)
• GRAVITY-wide likelihood now works regardless of the number of spectral channels used
• Introduce experimental FixedPosition "orbit" parameterization for modelling eg image or interferometry data at a single epoch, where there is no interest in the orbit parameters

v3.0.2

Octofitter v3.0.2

Diff since v3.0.1

• Typo fix in corner plot (thanks @gotten)
• Display chain number while intializing with Pigeons
• Remove arbitrary limit on the number of jitter and offset terms in StarAbsoluteRVLikelihood (thanks @langfzac)

Merged pull requests:

• Allow arbitrary RV jitter and offset terms (#32) (@langfzac)

Closed issues:

• Number of RV instrument jitter/offset parameters (#26)
• Typo in axis label OctofitterPairPlotsExt.jl (#33)

v3.0.1

Octofitter v3.0.1

Diff since v3.0.0

Hot fix: rename HiparcosIADLikelihood to HipparcosIADLikelihood

v3.0.0

Octofitter v3.0.0

Diff since v2.5.0

• Overhauled plots
• Fit relative RV between a planet and star
• Fit unbound hyperbolic orbits
• Model Hipparcos IAD directly
• Rigorous 3D propagation of high RV / proper motion stars (Thanks Eric Nielsen!)
• Compute point wise likelihoods and cross-validation scores after sampling a posterior
• Use kernel density estimates as priors
• More automatic data sources for astrometry, RV, and atmosphere models
• Import/export posteriors in Orbitize! format
• Improved tempered sampling defaults (use stabilized variational PT by default)
• Bug fixes

Merged pull requests:

• CompatHelper: add new compat entry for OptimizationOptimJL at version 0.3, (keep existing compat) (#25) (@github-actions[bot])
• CompatHelper: add new compat entry for Makie in [weakdeps] at version 0.21, (keep existing compat) (#28) (@github-actions[bot])
• CompatHelper: add new compat entry for PairPlots in [weakdeps] at version 2, (keep existing compat) (#29) (@github-actions[bot])

v2.5.0

Octofitter v2.5.0

Diff since v2.4.0

• Fixes to @System and @planet macro scoping
• Improved error messages for some kinds of model mis-specifications
• New feature: load and save basic chains in Orbitize! HDF5 format.
• New feature: load posteriors and sample data directly from whereistheplanet.com
• New feature: compute point wise likelihoods and cross-validation scores via PSIS-LOO
• Tweak to Sine() prior for numerical issues (limits are now eps() instead of zero)
• UniformCircular() priors now have a less aggressive prior on the internal vector length
• Pigeons sampler now starts from a better location upon initialization
• Change: Enzyme is no longer loaded by default, and must be imported by the user. As a result, the package load time is much faster! The default :ForwardDiff autodial backend is unchanged