Migrating from 1.x to 2.0

Version 2.0 removes some arguments, rejects some inputs that 1.x ignored, and changes some results. No public name was renamed. 1.x code that uses a removed argument, attribute or import fails with a TypeError, AttributeError or ImportError, and a rejected input raises ValueError. Each table maps the 1.x usage to the 2.0 usage. The changelog lists every change.

Results from 2.0 differ from those in the paper (Speagle et al. 2025, arXiv:2503.02227). See Results that change.

Fitting

1.x usage	2.0 usage
fit(..., subsample_mode="topk") or any subsample_mode	Remove the argument. The max_models cap keeps each model with probability min(1, c * w) and weights it by the inverse.
fit(..., ltol_subthresh=...)	Remove the argument. Each model checks its own convergence.
fit(ltol=3e-2, logl_initthresh=5e-3), the old defaults	The defaults are ltol=1e-3 (the convergence tolerance) and logl_initthresh=1e-6 (the first cut; init_thresh in loglike_grid). Passing the old values does not restore 1.x results, because the optimizer changed.
max_models=0 to keep every model	Pass max_models=None. Other values must be positive integers.
logpost_grid returns 8 values	It returns 9. The ninth is ln_incl, the log-probability that the max_models cap kept each model.
fit(...) on a grid without agewt, such as the Bayestar grid	Pass apply_agewt=False. The default apply_agewt=True raises ValueError there.
fit(..., apply_dlabels=False) had no effect	It now evaluates the Galactic prior without model labels.
loglike_grid(..., unknown_keyword=...)	Unknown keywords raise TypeError. Fix the spelling.
An integer 0/1 data_mask, which 1.x read as band indices	Nothing to change: it now masks bands, so the results change. Boolean masks give the same results as before.
parallax without parallax_err	Pass both, or neither. One alone raises ValueError.
An object with no valid bands stopped fit	fit skips it and writes model_idx = -99 and obj_Nbands = 0.
Zero probability returned -inf	It returns LOG_ZERO (-1e300) in brutus.utils.math.

Priors and dust

1.x usage	2.0 usage
logp_extinction(..., avlim=None)	Pass the A(V) range of the fit, for example brutus.utils.math.AVLIM. None raises ValueError.
logp_extinction without scatter	It adds scatter=0.2 mag in quadrature to the map spread. Pass scatter=0 to remove it.
logp_extinction(dustmap) with a 3D map and no distance	Pass distance. It raises ValueError otherwise.
Bayestar.get_query_size(coords)	Remove the call. The method no longer exists.
brutus.dust.__version__	Use brutus.__version__.
logp_galactic_structure with labels of the wrong length	Pass one label per distance. It raises ValueError.

Stellar models and grids

1.x usage	2.0 usage
GridGenerator.make_grid(dist=...)	Remove dist. Grids are always at 1 kpc.
make_grid with a one-value av_grid or rv_grid	Pass at least 2 distinct values for each.
StarEvolTrack.get_seds(..., Av=5.0), or any unknown keyword	Use the documented argument names. Unknown keywords raise TypeError, also in StarGrid.get_seds and get_predictions.
StarGrid.get_seds with a label outside the grid	Pass a label inside the grid. It raises ValueError.
StellarPop.get_seds(return_dict=False) returned params of shape (Npred, Neep)	params has shape (Neep, Npred), like params2. Transpose old code.
StellarPop.get_seds fell back to per-star SEDs when sed_batch failed	The error propagates.
RuntimeError from StarEvolTrack or Isochrone.get_predictions	The original error type passes through. Catch that type.
EEPTracks.null, EEPTracks.binwidths	Remove the attributes. They were unused.
from brutus.core import _get_seds	Use from brutus.core.sed_utils import _get_seds.
Old EEPTracks caches (*_cachev2.pkl)	The first load builds a *_cachev3.pkl file. Delete the old file.

Data

1.x usage	2.0 usage
fetch_isos("./data")	fetch_isos(target_dir="./data"). The file choice comes first. target_dir is keyword-only for every fetch_* function.
fetch_offsets("./data", "mist_v9")	fetch_offsets("mist_v9", target_dir="./data")
load_models without an afe or smf column	The columns are present as one-value columns when the grid is constant in them. Do not index by column position.
load_models(labels=['mini', 'feh'], include_postms=False) kept every model	The cuts apply whatever labels holds. A phase cut on a file without eep raises ValueError.
load_models(include_ms=False, include_postms=False)	It raises ValueError.

Offsets

1.x usage	2.0 usage
analysis.photometric_offsets(..., dim_prior=...)	Remove the argument. The function always uses Gaussian leave-one-out weights.
plotting.photometric_offsets(..., dim_prior=...), also photometric_offsets_2d	Remove the argument.
PhotometricOffsetsConfig(random_seed=seed)	Pass rng=np.random.default_rng(seed) to photometric_offsets.
PhotometricOffsetsConfig(use_vectorized_bootstrap=..., validate_inputs=..., progress_interval=...)	Remove the arguments. Input validation always runs, and verbose controls progress messages.
sel as an index array, which 1.x accepted and then excluded the wrong objects	Pass a boolean or 0/1 array of shape (Nobj,). An index array raises ValueError.

Plotting

1.x usage	2.0 usage
hist2d(..., unknown_keyword=...)	Remove the keyword. It raises TypeError, also for an unknown key in cornerplot(hist2d_kwargs=...).
cornerplot(applied_parallax=True)	The default is None: the parallax applies when parallax and parallax_err are both given.
dist_vs_red and bin_pdfs_distred with the A(V) range 0 to 6	The default range is 0 to 20 mag. Pass span for a narrower range.
smooth=0.015 in dist_vs_red, smooth=0.01 in bin_pdfs_distred	The defaults are (0.015, 0.005) and (0.01, 0.005). Pass the old scalar to keep the old smoothing.
summary_plot with the global numpy random stream	Pass rstate.
dist_vs_red always drew on the current axes	It draws on ax from fig=(fig, ax) when given.

Package

1.x usage	2.0 usage
Python 3.9, numpy below 2.0	Use Python 3.10 or later and numpy 2.0 or later.
Building from source with an old setuptools	Use setuptools>=77.
import brutus with a missing dependency warned	It raises the original ImportError.
brutus.__version__ from a source tree	Install the package. An uninstalled tree reports "unknown".

Results that change

The changelog lists each change in results for the same inputs and seed. The entries marked (output-changing) change fit results, and most of them are in the first “Fixed” section under 2.0.0. Other entries change plotted values and the log-likelihoods of the cluster and line-of-sight fits.