- All plots are downloadable using the download button.

- Hurdle models added as
`HP`

and`HNB`

options for the`dist`

argument. - The weighted and robust options removed from the Shiny app UI.
- Hurdle model based PI calculation implemented with 0-truncated P and NB distributions.
- The new
`loo`

function calculates the leave-one-out error as blended Chi-square distance in`$chi2`

. - The blended Chi2 based GoF metric is added to the
`mc_models_total`

table.

- Shiny app improvements (#21): Crosstalk-like behavior for univariate plots, Histogram binning, Total Prediction map/summary subsets, Export data checks.

- Add option to remove intercept(s) in regression to allow regression through the origin.

- Add robust regression option.

- Bootstrap setting is now a numeric input, not a slider, to allow high numbers.
- Composition analysis can be performed by using the cell level stats and by reusing an existing PI object.
- CPI calculations do not exclude rows with unknown ages.
- CPI summaries return Mean and Median besides PI.
- Unknown age classed individuals are distributed randomly according to comp model.
- The Shiny app uses the PI object when available.

- Added Residual tab for composition analysis to show model summaries and AIC table.
- More columns are allowed to be treated as filters/subsetters with 2 levels (usually 0/1).
- Print abundance/density summaries for subsets under Summary tab.
- Added
`Total_Bulls`

and`Bulls_per_Cow`

to the composition summaries. - Yearling cows now calculated as
`min(adult_cows, yearling_bulls)`

to avoid negative mature cows values. - Total Calves / Total Cows is now part of the
`mc_summarize_composition`

output.

- Composition table contains BIC and coefficients.
- Composition checks fixed: don't include unknown ages.
`mc_plot_univariate`

has the ability to return ggplot2 objects and their interactive versions.- New function:
`mc_plot_predfit`

to check model fit. - Shiny app: predictions can now be subsetted independent of the training data.

- Composition analysis added to R package and Shiny app.

- Total moose estimation is tested and ready to be used in the field.
- Added
`run_app()`

function to launch a Shiny app.

- Package renamed from DeducerPlugInMoose to moosecounter, with the intention of dropping the rJava/Deducer GUI features in favor of a Shiny app. Version numbering is continuous.

- Composition analysis fix: vglm failed when input matrix rowsum was 0. Now it is treated as missing and omitted.

- Added
`nobs.zeroinfl`

method.

- Fixing Non-ZI model summaries.
- Increment version.

- Non-ZI versions of count models (P and NB) added by introducing a hacked version of zeroinfl: zeroinfl2.

- Global option
`wscale`

added to tune weighting scale.

- Dual (weighted & unweighted) prediction added to composition PI calculations.

- Dual (weighted & unweighted) prediction is performed depending on survey area (unsurveyed gets predicted under unweighted model to better represent the high end of PI, surveyed area gets predicted under weighted model because high values are already captured within surveyed cells).

- Write down PI/CPI algorithm
- Accumulate issues and report as part of the summaries

- Error catching fixes in CPI calculation.

- Error catching fixes in PI calculation.

- Exposed optim method through options, might be a good idea to set it to Nelder-Mead (because of weighting instability using BFGS).

- Fixed issue with
`plotResiduals`

and`plot_predPI`

: failed when there were no outliers with error`zero-length 'labels' specified`

.

- PI distribution plot is tweaked to display sensible results for the case when the distribution has only one unique value (i.e. 0s).

- ZIP added as option beside ZINB (univariate exploration and model fitting).
- Weighted modeling option added to minimize influence on predictions.
- Histograms show % instead of density (%=100*density).
- Residual plot labels +/- 1.5*SD points and uses symmetric divergent coloring.

- Cell level stats now include Mode as well.

- ZI prediction did not use covariate coefs, only the intercept that led to biased simulations.
- Model averaging now uses selected models from dialog instead of all model in the ModelTab
- Explicitly imports fitted and model.frame methods from VGAM for composition analysis to avoid scoping issues with formula
- Number of hist bins in PlotPiDistr can be set by the user.
- Added AIC weights to composition model table.
- Composition PI now has the option for model selection similar to total Moose PI.

- Fixed alpha level: was not available for some functions from options.
- Null ZINB model in model dialogue fixed (length=0, not =="")

- Allow ZI to vary with covariates, UI and help dialog updated. PI simulations how use the model with ZI covariates as well.
- Model averaging function tracks the fitted values and uses the model averaged mean as the fit.
- Started adding Rd files with function documentations.

- Added multi-model averaging to PI simulation: use weights from model table to select models to refit and the option is added to the UI dialog.
- Checked and updated help dialogues to reflect updates.
- Added mode to total moose PI table.
- Added plot of PI distribution (PI) to UI: called 'Plot pred. distribution' in the menu, option to select full or subset PI data.
- pb option set to "none" (tcltk froze Mac).

- Bootstrap mean with 2 decimal places added to PiData table.
- Unknown animals are dropped from compositional analysis with a warning.
- Sightability correction can be defined up front, defaults to 1.
- Summary tables print decimal numbers instead of scientific notation.
- Total adult cows can be used as response for total models.
- Calf/cow ratio and related compositional analysis added.

Incremental production versions.

Initial release