Distance 7.3 Release 2
Posted 07 July 2020 by Len Thomas
We are pleased to announce an update to our Distance for Windows software: Distance 7.3 Release 2. The new release updates the R packages used by the MRDS, DSM and MA analysis engines, and the simulation engine, with bug fixes and compatibility with R 4.0. If you use any of these engines, we recommend you download the new release.
The new release contains versions of the mrds, dsm, mads and DSsim packages, which were compiled under R 4.0.2. They implement the following changes:
- fix issue in predict() when uniform key functions are used with new data.
- new function p_dist_table() to show the distribution of estimated probabilities of detection. Useful for covariate models to determine issues with very small ps.
- new function add_df_covar_line(), which can be used to add lines plots showing the detection function for a given covariate combination. Thanks to various members of the distance sampling mailing list for this suggestion.
- plots produced by plot.ds/plot.rem/plot.rem.fi/plot.trial/plot.trial.fi/plot.io/plot.io.fi/plot.det.tables now use same defaults as R 4.0.0 (“lightgrey” bars for histograms). Some deprecated arguments to plot.ds were removed.
- hessian now returned when solnp (constrained optimisation) is used to fit the detection function
- Check for NA covariate values, thanks to Ana Cañadas for highlighting this issue.
- enable P2 variance estimator for points
- Corrected handling of NA covariates
- new option se.fit for predict.ds to obtain standard errors for the probability of detection or ESW
- Fixed a bug in dht when left truncation is used. Previously left truncation was ignored. See https://github.com/DistanceDevelopment/mrds/issues/22 thanks to Carl Schwarz for finding this bug.
- Fix bug where two objects could have a missing observer and no error was thrown. Thanks to Ainars Aunins for reporting this bug and Eric Rexstad for diagnosing.
- transect type (line or point) is now determined from the detection function rather than using the transect= argument
- multiple detection functions are now possible, use a list() of fitted detection functions supplied to ddf.obj= argument to dsm(). dsm.var.gam, dsm.var.prop and dsm_varprop are compatible. Vignette coming soon. Thanks to Dave Fifield and Ewan Wakefield for extensive testing and suggestions.
- strip transects now handled using the dummy_ddf() function.
- fixed dsm_varprop (gam.fixed.priors) for response distributions with fixed scale parameters. Negative binomial results before now may have been incorrect!
- Clarification of the predict.dsm return documentation. Thanks to Dave Fifield for the suggestion.
- off.set=NULL lead to predictions of 0 when density was used. Now just returns density. Thanks to Dave Fifield for the catch.
- Now Suggests package “sp” (needed for some parts of the mexdolphin example data)
- Added % complete progress counter
- Fixed example dataset
- Fixed warnings bug generated when there were too few data points
- Code to calculate perpendicular distance is more robust
- Fixed incorrect true expected cluster size values in simulation summary
- Documentation corrections
- Can now generate equal spaced zigzag designs with any design axis angle and can generate two sets of complimentary zigzags.
- User can now select between using all repetitions where at least one model converged and ony repetitions where all models convgered. This is selected when calling summary on a simulation object.
- Now tells user which models cause issued in mrds.
- Improved backwards compatibility, including updating the S4 objects to include default values for missing slots.