gadget3 1.0-0:

The 1.0 release coincides with the publishing of the introductory gadget3 paper “gadget3: An evolution of the gadget modelling framework”: https://doi.org/10.1016/j.fishres.2026.107903.

gadget3 0.15-0:

Bug fixes

g3_tmb_relist() was double exp()ing fixed parameters #260
Support multiple areas in g3a_age() #262

New features

Configurable scale for g3l_bounds_penalty() #263

Breaking changes

Re-work interim_value to be per-fleet, parameter by default #264

gadget3 0.14-0:

Bug fixes

g3a_predate_catchability_project now correctly scales when quotas are expressed in individuals #222
gadget3 resistant to packages (re)defining head/tail S3 methods, e.g. reformulas #224
g3_param_project projections tending to zero in some cases #226
g3_quota_hockeyfleet not overreporting when handed multiple stocks #227
Allow for zero mean weight in weightjones #228

New features

Add g3a_trace_var, based on g3experiments::g3a_trace_nan #231
Add g3a_trace_timings, for investigating relative action speed #235
Add g3_tmb_fn(), for projection runs & similar #237
Parameters can be set as type=“LOG”, for optimisation in logspace #251

Breaking changes

optimise defaults to FALSE unless default bounds are provided #233
Don’t generate TMB ADGrad function if random effects not used #234
Hide g3a_initialconditions / g3a_renewal (not g3a_initialconditions_normalparam/cv) #240
g3_param_project_dlnorm() / _logar1() accept parameters in linear space, not log #251

gadget3 0.13-0:

Bug fixes

All-zero lengthgroups in initialconditions / > 1 maturity ratios no longer result in NaN #187 #182

New features

The stock column in likelihood data can now contain name parts as well as full stock names
g3_to_r-produced model functions will accept both a list & data.frame parameter template
g3_to_r models can include arbitary R functions, see vignettes/writing_actions #194
Add g3_array_agg, g3_array_combine #193 #213
Add surplus production model action g3a_spmodel #209
Projection / MSE support #210 #220
- Add g3_param_project() for time-varying parameters that also support projection
- AR1/logAR1 distribution functions
- Add g3_quota() for quota-definitions during projection based on a custom fishing year
Add g3_array_plot

Breaking changes

run_projection is now TRUE by default for renewal. A rec.proj parameter covers projection.
Rename detail_x__num to dstart_x__num #200

gadget3 0.12-0:

Bug fixes

g3a_spawn() splits offspring into multiple stocks correctly

New features

Split likelihood distribution by predator_length / predator_age / predator_tag #154
Predation support, see ?g3a_predate / ?g3a_predate_catchability_predator #29
g3a_otherfood, g3a_otherfood_normalparam, g3a_otherfood_normalcv #155
Support length transform_fs in g3l_catchdistribution #95

gadget3 0.11-0:

Bug fixes

g3_to_r() output can be debugged in R-Studio
Unknown stock names in likelihood data now an error
g3l_distribution_surveyindices_*() can now be broken down by age without error

New features

Add g3_parameterized('x', by_year = 1998:2099) to override year range for parameters
Add g3a_initialconditions_normalcv for initialconditions driven by a lencv parameter
Add g3a_renewal_normalcv for initialconditions driven by a lencv parameter
by_predator parameter for g3_parameterized
Defaults for g3_suitability_exponentiall50
Support cut() formatted character columns as well as factor columns in likelihood observation data.
Add g3_init_val() helper, to replace gadgetutils::g3_init_guess()
Add g3_areas() helper to create numbered area vectors
Add g3_distribution_preview() helper to see how observatation data will be converted to array
period now optional for g3l_random_walk(), g3l_random_dnorm()
g3l_bounds_penalty() can now generate bounds from actions which update automatically

Breaking changes

g3l_distribution_surveyindices_* now default to beta = 1, set beta = NULL to restore previous behaviour.
stock_param() & stock_param_table() replaced with stock_prepend()
Remove unused g3_to_tmb(adreport_re) option #60
Naturalmortality M now by-age by default #113
g3_suitability_andersenfleet() now uses by_predator = TRUE by default #133

gadget3 0.10-0:

New features

Add g3_parameterized('x', by_step = TRUE) for seasonal parameters #115
Support R factor levels as well as MFDB-style attributes for likelihood observation data. This means dplyr group_by(length = cut(...)) can be used as well as MFDB for groupings #112
Parameterised defaults for g3l_understocking(), g3a_naturalmortality(), g3a_renewal_normalparam()
Add g3a_renewal_vonb_t0() & g3a_renewal_vonb_recl(), with g3a_renewal_vonb() being an alias to the latter

Bug fixes

g3s_modeltime() (used for history) now works when final_year_steps set
Stop assuming 1-dimensional R arrays should be TMB vectors.

Breaking changes

Remove scale = 0.001 from default K parameters in g3a_grow, g3a_renew_*
g3a_renewal_vonb_t0() is now the default mean_f for *_normalparam()
g3a_initialconditions_normalparam() now offsets any age in mean_f (i.e. the VonB formula) by cur_step_size, in effect running at step -1.

gadget3 0.9-0:

Bug fixes

g3_suitability_andersen() now produces sensible values - https://github.com/gadget-framework/gadget3/issues/108
g3a_age() now supports stocks with a single age (i.e. minage == maxage)

New features

Add recage parameter to g3a_renewal_vonb / g3a_renewal_initabund
Add g3_suitability_andersenfleet(), a fleet-specialised andsersen suitability function.
Make g3_is_stock() public
g3_eval(), to evaluate snippets of a gadget3 model.

Breaking changes

g3l_distribution_*(transform_fs = ...) now happens before aggregation, not after. Any matrix used now has to be expressed in terms of the stock, not aggregated age.
g3_param_table() now returns NaN (and warns) on a missing value, instead of aborting.
optim(g3_tmb_par(...)) is now an error. When optimising, always use obj.fun$par.

New features

Support as.vector(array) in TMB formulas, so arrays can be used with TMB vectorized functions, e.g. pnorm(as.vector(ar[,1])).

gadget3 0.8-0:

New features

Initial CRAN release