bayesnec.step function to toolkit to allow for
brms non-linear formula evaluation (89edef72).Fixed issues with summary and print
methods (bf058f36).
Plotting methods now display the proper no-effect toxicity legend label (2be55d51).
Implemented native brms::beta_binomial() instead of
custom family beta_binomial2 (c6c78fb7).
compare_* functions now have N(S)EC as default
behaviour (c4c84ba1).
Changed argument name precision to
resolution in many functions to maintain syntax
compatibility with brms (727d6700).
Standardised plot style between sample_priors and
check_priors (6c59af07).
Improved speed in test runs for CRAN (d5d097fb).
Aligned generic method consistency for rhat (afb10577).
Fixed a bug relating to 01 bounded x data in
define_prior (3ab93e0).
Fixed messages returned for set_distirbution (f483160).
Reduced n_trials and deleted unnecessary set.seed in while call
for make_good_inits (3c5f084).
Fixed bug introduced when implementing recycling the seed passed
to bnec (01394a17).
Included additional criteria in the initial values algorithm to ensure initial values can be fit in stan (fe89484a).
Cleaned parameter title names in sample_priors (6973bae1).
Small tweak to initialisation search, such that the seed used for
a bnec call (via ... arguments to
brms::brm) gets recycled and therefore it generates the
same initialisation values across platforms (c74a3c46).
brm (30eb8e6).brms (c5c16be).gsub substitution of x in original
brms non-linear formulas, and added various tests (fa588fb).162fec9).Fixed CRAN bugs related to testthat (cfe9f20).
Changed decline detection behaviour in function
check_data (c2744e3).
Addition of herbicide data (2b13ef5).
streamlined check_* functions for negative
predictors vectors – no error anymore (via check_data), now
drop models with informative message (via check_models) (f134311).
bnec now works exclusively with formulas (4bccc07,
d3531b4,
bccb0da,
340269e).
Added +, c and update
methods via a common class bnecfit (d5e81ab,
3adf3fa).
Fixes to ecx, nec and nsec
following new formula-based implementation (6ec1372,
0a9d307
and 19a37b9).
Fixes to predict methods following formula-based implementation
(8843a1c,
08484a7).
Dispersion now takes a bayesnecfit as input (200bc11).
Added brms::bernoulli to the list of allowed model
distributions (7606461).
Made identity the default link (6640867).
Added more controls and tweaks to checking model relative to
input data (85a3819),
065a445).
Added formula and model.frame methods
to main classes (9df4a64).
Upgraded ggbnec to autoplot as
bayesnec standard ggplot2 plotting method (65fe15f).
Enhanced handling of argument loo_controls in
bnec to allow argument for both
brms::add_criterion and loo::loo_model_weights
(9712fcc).
Added Bayesian R2 to summary (9a71a3c).
Expanded bnec’s capacity to accept input x and y
vectors, data frames (f256b39)
and formulas (32e74ac).
Corrected error for logit link cases for the beta_binomial2 where
data contain 0 and 1 to ensure appropriate prior values on top and bot
(4158237).
Series of internal fixes to standardise function class outputs
(81369bb,
1c70efe,
5bce2b5,
455ca70,
5e6b41e).
If link functions are not specified in bnec, then the default link function is used; previous versions of bayesnec used the identity link.
An additional family has been added betabinomial2 for over dispersered binomial data.
The package supports using link functions for generalized modelling, which appears to be more stable and is also in line with more typical generalised modelling approaches.
There are multiple options for model weights calculation from the loo package. The default is “pseudobma BB”.
There is now a compare_posterior function that also includes a bootstrapping procedure. This can be used to compare model fits across different datasets, or even different model sets for the same dataset (ie nec v ecx models). Please see the vignette for examples of usage.
There is a vignette detailing the models available in bayesnec. Note that not all models are suitable for all families, and also depending if link functions are used.
A new check_chains function has been added to allow chain plotting in base R and that works more smoothly with plotting chains for multiple fits for bayesmanec objects.