Home‎ > ‎

release notes

Revision history can be found here.

version 0.9-9

changes and new features
  • Added the new function, occuFP, which fits occupancy models in which false positive detections are possible. See Royle and Link (Ecology, 2006) and Miller et al. (Ecology, 2011). The new methods getB and getFP can be used to extract relevant information from the fitted models.
  • Added new function, gpcount, to fit the generalized N-mixture model discussed by Chandler et al. (Ecology 2011) to repeated count data collected using the robust design.
  • Changed the way model-averaged confidence intervals are computed using predict(fitList, ...). Rather than using the model-averaged SEs, the confidence intervals from the individual models are averaged. Note: the "level" argument now works.
  • non-parametric bootstrapping of fitted pcount() models now re-samples sites only.
bug fixes
  • Fixed getP method for models fit using gdistsamp(..., keyfun="uniform")
  • nonparboot now works with models fit using pcountOpen.
known bugs
  • Known bugs are same as for version 0.9-0 (see below), plus distsamp crashes on some Unix and Mac systems when engine="C". Use engine="R" if you encounter this problem. Hopefully all these issues will be fixed in version 1.0-0.

version 0.9-8

changes and new features
  • distsamp is now much faster because the likelihood is written in C++. The old slow version is still available by setting engine="R"
  • Added 'level' argument to predict, defaults to 0.95 = 95% CI.
  • Added new datasets: 'Switzerland', 'cruz', 'crossbill', and 'issj'. See their help pages for more information
  • Updated the spp-dist vignette. Now includes a distance-sampling analysis of the scrub-jay data
bug fixes
  • predict() method now returns the correct predictions of lambda when mixture="ZIP", but it currently does not return SE/CIs. This will be fixed eventually -- in the meantime, the bootstrap can be used.
  • ranef() method now works for pcountOpen(.., dynamics="trend")

version 0.9-7

changes new features
  • predict now works when newdata is an object of class RasterStack from the raster package
  • formatDistData now works for data from replicated surveys, and returns a matrix suitable for use with gdistsamp
  • added vignette on creating species distribution maps
bug fixes
  • fixed simulate and ranef method for models fit using gmultmix or gdistsamp with numPrimary=1.
  • fixed getP method for gdistsamp
  • pcountOpen would crash in an unusual situation involving many missing values. Fixed now.

version 0.9-5

changes and new features
  • Added empirical Bayes methods to estimate posterior distributions of latent variables (abundance or occurrence). See help(ranef). 
  • Added vignette on fitting capture-recapture models
  • pcountOpen is 2-5x faster now for "constant" and "notrend" dynamics
  • Added bracket subset methods for unmarkedFrameGMM, GDS, and PCO, which make it possible to use nonparboot on models fit using gmultmix, gdistsamp, and pcountOpen.

bug fixes
  • The bracket subset method for unmarkedMultFrames was broken when yearlySiteCovs were present. This caused nonparboot to crash. Fixed now.

version 0.9-4

bug fixes
  • The transition probability matrix was not being computed correctly in pcountOpen when "time gaps" were present, ie if there were missing data for an entire primary period at a particular site. Effect should have been minimal in most cases. Thanks to Jeff Hostetler for catching this.
  • colext, gdistsamp, and pcountOpen were not handling certain types of missing values correctly. An error was being thrown when it shouldn't have been.
  • contstant used to avoid log(0) was not small enough in pcountOpen. Now occu, pcount, and pcountOpen use DOUBLE_EXP in C code.

version 0.9-3

changes and new features
  • pcount and pcountOpen now allow for zero-inflated Poisson distribution (mixture="ZIP")
  • pcountOpen is faster with likelihood written in C++
  • pcountOpen has a new population dynamics model, dynamics="trend" (thanks to Jeff Hostetler for the suggestion)
  • gmultmix now sets phi=1 if T=1
  • additional arguments (eg control settings, lower, and upper bounds) can be passed to optim using the "..." argument of the fitting functions
bug fixes
  • occuRN now returns AIC instead of AICc-like criterion
  • pcountOpen accepts offsets (previously they were ignored without a warning)

version 0.9-2

changes and new features
  • Added reference to new paper describing unmarked. This should be used to cite the package. Type citation("unmarked") for details.
  • occu() and pcount() now have an argument "engine" allowing users to try a faster method of evaluating the likelihood. engine="C" should be 30-50% faster than engine="R".

version 0.9-1

changes and new features
  • Added new function, gdistsamp, to fit the model of Chandler, Royle, and King (2011) to distance sampling data. This allows g(0)<1 and negative binomial distribution for Ni
  • Added new vignette for colext. Thanks to Marc Kery for doing most of the work
bug fixes
  • Fixed bug in predict method for gmultmix
  • Fixed bug in simulate method for pcountOpen
  • colext now returns AIC instead of AICc
  • Known bugs are still the same as below (version 0.9-0)

version 0.9-0

changes and new features
  • Added the function pcountOpen() to fit the model of Dail and Madsen (Biometrics In press). See help(pcountOpen).
  • unmarked now depends upon the Rcpp and RcppArmadillo packages.
  • Added getP and simulate methods for colext fits so that parboot can be used.
  • distsamp() reverts to an approximate integration when integrate() fails. This usually happens when the detection function is flat.
  • Increased the default value of K used by pcount and gmultmix.
  • Added help pages for each unmarkedFrame type describing how to format data.
  • Added more quality control tests to R_home/inst/unitTests
bug fixes
  • yearlySiteCovs were not being copied correctly to obsCovs during creation of design matrices used by colext().

known bugs
  • colext fails when there are missing yearlySiteCovs
  • formatMult converts numeric vectors to character vectors if factors are present in df.in
  • predict fails when newdata contains factors that do not have exactly the same levels as the original covariate used in the analysis. There is a simple workaround described in this thread.
  • predict does not properly handle scale in model formula.

version 0.8-9

changes and new features

  • Added more tests to inst/unitTests
  • Added example to ?unmarked
  • logLik method is now available to the user. It returns a simple scalar in contrast to method in stats package.
bug fixes
  • Model order is no longer important in LRT
  • Corrected model-averaged SE calculation in predict method for unmarkedFitList (thanks to Marc Mazerolle)
  • Fixed multinomial cell probability calculation in distsamp for survey="point".

version 0.8-8

changes and new features
  • now depends on >= R-2.12.0
  • summary method for unmarkedMultFrames includes yearlySiteCovs
  • predict methods for colext and fitList return CIs
  • formatDistData returns matrix instead of data.frame
  • added new unit tests in R_HOME/inst/unitTests
  • added rel.tol argument to distsamp to provide some control of integrate function
  • added coef and SE methods for fitList objects
  • removed n from modSel show method
bug fixes
  • fixed lambda offset of gmultmix
  • fixed simulate method for gmultmix
  • fixed cn() so that modSel won't break with singular hessians
  • fixed NA handling for gmultmix
  • fixed doublePiFun
  • fixed bracket subsetting methods for unmarkedMultFrames