Changes in version 4.1.0 (2019-12-17)

New features or changed behaviour of old functions
    o   AICcompatible() new S3 method to check models for AIC compatibility
    o   AIC() warns if secr models not compatible
    o   LR.test() stops if secr models not compatible
    o   secr.fit() faster fitting of polygon and transect data
    o   secr.fit() option details = list(convexpolygon = FALSE) allows non-convex polygons.
    o   autoini() uses simpler and more robust code; secr.fit() start values differ slightly 
        from earlier versions; new argument 'adjustsigma'.

Bugs fixed
    o   reduce.capthist() subtracted dead animals when outputdetector = "count"
    o   verify.capthist() spurious warning for transect data with character animalID
    o   derived() bad for polygon and transect data in secr 4.0 because esa() corrupted 
    o   secr.fit() in secr 4.0 bad likelihood and estimates for some covariate models 
        (bad indexing of parameter array PIA)
    o   secr.fit() failed with multi-session mark-resight data (fixed pID, non-null Tm)
    o   suggest.buffer() failed with input from multi-session fitted model

Other
    o   RcppNumerical used for numerical integration with polygon and transect detectors
    o   requires R >= 3.5.0 for obscure reasons (serialization of data objects)
    o   suppressed new warnings from regularize.values in approx()
    o   internal function secr.design.MS() argument bygroup replaced with CL
    o   LLsurface() returns value matrix when plot = FALSE (otherwise invisibly as before) 
        (NA where likelihood could not be evaluated)
    o   mark-resight sighting only, partial identification re-enabled

Changes in version 4.0.2 (2019-11-11)

New features or changed behaviour of old functions
    o   secr.fit() LLonly = TRUE stores number of parameters and timing information
    o   secr.fit() LLonly = TRUE uses default or autoini start values if none provided
    o   summary.secr() versiontime includes elapsed time
    o   summary.capthist() includes summary of individual covariates

Other
    o   CRAN release of 4.0

Changes in version 4.0.0 (2019-10-27)

New features or changed behaviour of old functions
    o   'ncores' argument defaults to NULL in most functions (meaning one less than number
        of available cores for multi-threading)
    o   fx functions 'normal' argument dropped
    o   'ncores' argument dropped from mask.check(), pmixProfileLL() and simulate.secr()
    o   autoini(), region.N() and suggest.buffer() acquire 'ncores' argument
    o   fast secr.fit() for binary and count proximity detectors (default details$fastproximity = TRUE)
    o   make.systematic() new argument 'order' to control cluster sequence
    o   RSE() extract precision of estimated parameter from fitted model
    o   par.secr.fit() no longer automatically sets ncores=1 in the separate calls to 'secr.fit'
    o   secr.fit() details option autoini = 'all' for start values from multi-session data
    
Bugs fixed
    o   sim.secr() with 'multi' detectors slightly understated detection rate owing to mis   
        -assignment of detection locations 
    o   make.systematic() and trap.builder() ignored region polygon when cluster not
        specified (edgemethod set internally to 'allowoverlap'

Other
    o   converted C code to C++ with multithreading in RcppParallel as appropriate
    o   some features of earlier versions have yet to be implemented in 4.0
    o   secr object from secr.fit() includes design matrices for D and noneuc
    o   functions read.SPACECAP() and write.SPACECAP() removed

Changes in version 3.2.2 (2019-07-02)

New features or changed behaviour of old functions
    o   summary method for popn objects
    o   write.mask() function

Bugs fixed
    o   sim.popn() exponential
    o   plotMaskEdge() sometimes shows internal lines (non-integer spacing)
    
Other
    o   write.capthist help wrote files to disk; now in \dontrun{}

Changes in version 3.2.1 (2019-06-01)

New features or changed behaviour of old functions
    o   sim.popn() allows exponential and t2D movement between sessions
    o   sim.popn() new details arguments movemodel, move.a; sigma.m deprecated

Bugs fixed
    o   make.capthist() with transect data gave error when points did not lie on transect
    o   moves() failed with nonspatial capthist (also summary.capthist(moves = T) etc.)
    o   plot.popn() ignored frame argument when collapse = TRUE
    
Changes in version 3.2.0 (2019-01-25)

New features or changed behaviour of old functions
    o   derivednj() extended for weighted variance (R2, R3), with subtle argument changes
    o   derivednj() and dependent functions return 2-line dataframe (rows 'esa', 'D') as for derived()
    o   derivedSession() extended for weighted variance
    o   derivedCluster() extended for weighted variance
    o   derived() for secr and secrlist objects acquires argument 'bycluster'
    o   split method for mask objects (split.mask)
    o   trap.builder() new edgemethod and exclmethod options "anyinside", "centreinside" etc.
    o   derived.secr() improved handling of sessions with zero detections
    o   derivedSystematic() implements Fewster 2011 variance for systematic cluster designs
    o   make.systematic() acquires argument "chequerboard", saves originbox, 
        argument 'originoffset' replaces 'bufferorigin'
    o   spacing.traps() returns numeric(0) instead of NULL for 1-trap arrays
    o   write.traps() removed single quotes around objectname
    o   secr.fit() allows mixture models when detector 1 unused on occasion 1
    o   make.mask() more explicit warning when nrow = 0
    o   plot.traps() gridlines = TRUE respected for add = TRUE (default gridlines = !add)
    o   pdot.contour() and related functions acquire argument 'poly.habitat'
    o   summary method for capthist objects (summary.capthist) optionally reports movements

Bugs fixed
    o   make.systematic() grid didn't span extremes of some region polygons
    o   sighting-only did not allow unidentified marked sightings
    o   reduce.traps() failed for reduced number of occasions when both usage and markocc specified

Changes in version 3.1.8 (2018-11-19)

New features or changed behaviour of old functions
    o   make.systematic() saves origin and centres as attributes
    o   make.systematic() new argument bufferorigin controls random origin
    o   summary method for secr objects, with print and AIC methods for summary (may change)
    o   clone.popn() rownames consistent with clone.capthist() (1.1, 1.2...)
    
Bugs fixed
    o   make.mask() not using keep.poly and check.poly arguments for multi-session masks
    o   detector usage not correctly included in some models with detector type "multi"
    o   clone methods not robust - freq now rounded to avoid quirky behaviour of rep()

Other
    o   some models (e.g., g0~bk with detector type "multi") fit more quickly than before

Changes in version 3.1.7 (2018-10-03)

New features or changed behaviour of old functions
    o   esa.plot() computes CVpdot for varying buffer width 
    o   esa.plot 'as.density' argument replaced with 'type' and new argument 'conditional'
    o   sim.popn() edgemethod new options 'stop' and 'reflect'
    
Bugs fixed
    o   OVpossumCH dataset Ageclass covariate placed 1-year-olds in >=2 year class
    o   join() detector attribute not expanded correctly for some inputs
    o   reduce.capthist() bug introduced in 3.1.6
    o   subset.popn() failed for multi-session populations
    o   region.N() gave misleading R.N for models with individual variation in detection, 
        including h2
    o   trap.builder() failed when projection specified for 'region'
    o   addTelemetry() failed with mark-resight data
    o   write.capthist() lost non-numeric detector names (since 2017-11-30)

Other
    o   secr-manual.pdf is no longer distributed with package - see 
    https://www.otago.ac.nz/density/pdf/secr-manual.pdf or https://CRAN.R-project.org/package=secr/secr.pdf.

Changes in version 3.1.6 (2018-05-14)

New features or changed behaviour of old functions
    o   sim.popn() details (turnover) argument "wrap" deprecated; use "edgemethod"
    o   plot method for capthist object copes with missing rownames
    o   CVpdot() new function for CV of p. across mask
    o   improved handling of attributes (intervals and sessionlabels) in subset and reduce
        methods for capthist objects
    o   join() allows site matching by name, which is faster
    o   timevaryingcov() extended to capthist objects for openCR - see ?join
    
Bugs fixed
    o   internal fn predictD returned matrix instead of vector, causing problems in region.N
    o   sim.capthist() failed for signal detectors (df0 error reported by Ben S)

Changes in version 3.1.5 (2018-02-25)

New features or changed behaviour of old functions
    o   rotate method for capthist objects
    o   secrlist() returns input secrlist object unchanged
    o   plotMaskEdge() accepts multi-session input
    o   secr.make.newdata() rewritten; acquires logical argument 'all.levels'
    o   predict method for secr objects passes dots argument to secr.make.newdata()
    o   join() intervals optionally passed as attribute 'intervals'
    o   coef method for secrlist objects
    o   unRMarkInput() assumes all freq = 1 if freq field missing
    o   new detectfn 19 (HVP hazard variable power)
    o   RMarkInput ignores 'covariates' argument when no covariates
    o   reduce and subset methods for capthist objects allow nonspatial input and output
        when verify = FALSE
    o   reduce and subset methods for capthist objects honour 'intervals' attribute
    o   summary method for nonspatial capthist objects drops detectors from table of counts
    o   plot method for popn objects acquires new arguments for overlaying multiple sessions
    o   sim.popn() argument sigma.m < 0 causes random relocation
    o   sim.popn() open population recruitment model "multinomial"
    o   LLsurface now a generic function with method for 'secr' objects
    o   new capthist attribute 'sessionlabels' useful for joined capthist
    
Bugs fixed
    o   secr.fit gave misleading estimates with binomial count data when size derived from usage 
        (reported by Mathias Tobler)
    o   predict.secr misreported pmix for other than simple h2 models (pmix = 0.5 for some levels)
    o   detectfnplot() in error when detectfn = 'HAN'
    o   sim.capthist() and sim.resight() failed in 3.1.3 when detector type varied by occasion
    o   make.poly() failed with some polygon list inputs

Other
    o   optional 'interval' attribute of capthist objects renamed 'intervals' (see ?join)
    o   export internal function insertdim()
    o   minor tweaks for compatibility with open population code
    o   NEWS for versions before 3.0.0 archived 
    o   some old documentation removed to online vignette secr-models.pdf

Changes in version 3.1.3 (2017-12-02)

New features or changed behaviour of old functions
    o   secr.design.MS() acquires argument 'contrasts' (used internally)

Bugs fixed
    o   sim.capthist() ignored recapfactor provided in detectpar

Other
    o   region.N now a generic method with functions for secr and secrlist objects
    o   corrected some obsolete documentation references to maptools package

Changes in version 3.1.1 (2017-11-30)

New features or changed behaviour of old functions
    o   new detector type 'capped' in make.grid(), sim.capthist(), reduce(), verify(), secr.fit() etc.
    o   as.data.frame methods for capthist and traps objects
    o   secrlist() and MS.capthist() generate names from input objects
    o   circular.r() and circular.p() acquire argument 'upper' for explicit upper limit of integration
    o   LR.test generalised to any fitted models with a logLik method
    o   plot method for secr objects: confidence limits now computed on log scale 
        for hazard detectfn HHN, HEX etc. to match predict.secr
    o   rbind method for traps adds suffix to names only when needed to avoid any duplication
    o   verify method for capthist objects checks for all-zero histories and missing rownames
    o   summary method for capthist objects: 'terse' option activated for single-session objects
    o   subset method for capthist objects: subset may be defined as function (useful for covariate-based selection)
    o   join() acquires arguments interval, to set between-session intervals directly, and sessions, for session names
    o   secr.fit() details argument 'contrasts' for specialised coding of factor predictors (example in secr-multisession.pdf)
    o   secr.fit() details argument 'newdetector' to override detector type
    o   'newdetector' reported by print.secr

Bugs fixed
    o   predict method for secr objects failed to report SE for parameters with some fixed coefficients (fixedbeta)
    o   verify.capthist failed to report detections at unused detectors

Other
    o   requires package stringr for some functions
    o   class attribute of multisession objects changed to place "list" last -- 
        c("list", "capthist") becomes c("capthist", "list")
    o   LLsurface.secr automatically substitutes lambda0 for g0 given detectfn 14:18
    o   experimental "nontarget" capthist attribute
    o   make.mask() explicit error when capthist passed 
    o   derived is now a generic method with functions for secr and secrlist objects
    o   derived.nj() renamed derivednj(), derived.mash() renamed derivedMash() etc.
    o   && changed to & throughout - should have no effect except fixes verify bug, but be wary

Changes in version 3.1.0 (2017-09-20)

New features or changed behaviour of old functions
    o   par.secr.fit() acquires arguments save.intermediate and LB to allow
        load balancing (thanks to Mathias Tobler)
    o   par.secr.fit() default value of 'seed' changed from 123 to NULL
    o   rbind.popn() now an S3 method; no longer accepts list input
    o   rbind.capthist() now an S3 method
    o   randomHabitat() acquires argument 'seed', saved as an attribute
        (thanks to Erin Peterson)
    o   detectfnplot() acquires argument 'hazard' for plotting hazard 

Other
    o   adjustment to 'ts' predictor for secr.fit mark-resight models
    o   documentation of random number seeds in ?secrRNG
    o   start vector returned by secr.fit now named

Bugs fixed
    o   faulty start vector when model included both h2 and noneuc

Changes in version 3.0.2 (2017-06-07)

New features or changed behaviour of old functions
    o   make.grid() and make.circle() acquire argument 'markocc'
    o   read.traps() accepts Excel xls or xlsx input (requires readxl)

Bugs fixed
    o   plot.secr misleading axis label 'Detection lambda'
    o   sim.resight() could fail with markingmask and predefined popn

Other
    o   imports file_ext from tools package
    o   attributes of output from sim.popn now include Nbuffer and D
    o   speed increase for "count" detectors
    o   miinor adjustments to sim.popn for compatibility R 3.4.0

Changes in version 3.0.1 (2017-05-03)

New features or changed behaviour of old functions
    o   'unresolved' sightings now stored in Tn attribute of capthist
    o   sim.resight() acquires argument 'unresolved'
    o   plot.secrlist() accepts colour vector
    o   new function as.mask() to coerce traps objects for plotting
    o   moves() acquires argument 'names' to control sort order (new default)

Bugs fixed
    o   sightingPlot() legend ignored fill argument

Documentation
    o   addTelemetry omitted 'single' and 'multi' from permitted detectionCH types

Other
    o   discretize() updated for 3.0
    o   native routine registration completed

Changes in version 3.0.0 (2017-04-11)

New features or changed behaviour of old functions
    o   3-D data structure used for trapping data
    o   updateCH() for easy conversion of old data
    o   multiple detector types may be used in one session 
        (point detectors only: single, multi, proximity or count)
    o   marking and sighting occasions may be interspersed
    o   rbind.traps acquires argument 'checkdetector'
    o   user-provided distance matrix may be session-specific
    o   esa 'noccasions' may be less than ncol(usage)
    o   function occasionKey() to embellish petal plots
    o   explicit handling of 'uncertain' sightings Tn
    o   plotMCP() now works with data from point detectors
    o   details$debug in secr.fit() now integer with multiple levels
    o   sightingPlot() new function
    o   plot.capthist() new option "centres"
    o   secr.fit() argument 'details' may be given as character value
    o   telemetrytype attribute of traps objects replaces details$telemetrytype
    o   new addTelemetry() function and revised handling of telemetry data
    o   addTelemetry() allows multi-session data
    o   secr.fit() etc. auto predictor 'tt' nontelemetry vs telemetry occasions
    o   addCovariates() acquires argument 'replace'
    o   c new method for combining secrlist objects
    o   make.capthist() discards repeat detections at exclusive detectors, with a warning
    o   models for polygon or transect detectors requre hazard detection functions 
    o   esa.plot() accepts secrlist as input (curves overlaid)
    o   par.secr.fit() acquires argument 'prefix'
    o   predict.secr() acquires argument 'realnames' to select a subset of 'real' parameters

Bugs fixed
    o   make.grid() did not assign polyID attribute for polygon detectors
    o   snip() and discretize() for transects dropped occasions with no detections
    o   sim.capthist() failed when 'userdist' supplied (non-Euclidean distances)

Other
    o   ovenbird dataset provided with detector type "proximity"
    o   esa() more robust to variation when noccasions specified
    o   new details option 'splitmarked' for resightings (default FALSE)
    o   usagePlot() argument occasion renamed occasions
    o   more compact 'call' component of secr.fit output
    o   discontinued warning when detectfn requires changed parameters (g0 -> lambda0)
    o   plot method for traps enables box around plot (bty = 'o')
    o   plotMaskEdge faster
    o   secr.fit() details savecall = FALSE suppresses call (default in par.secr.fit)
    o   new function shareFactorLevels() to force factor covariates to use the same levels
    o   rgdal::readOGR used instead of deprecated maptools::readShapeSpatial
    o   minor speed improvments for models without learned response
    o   ip.secr() faster default predictorfn
