murrayefford / secr

Spatially Explicit Capture-Recapture

secr

Spatially Explicit Capture-Recapture

This is the development version of the R package distributed on CRAN.

It may be installed using

devtools::install_github("MurrayEfford/secr")

Compilation of C++ code is required.

Please report bugs as Issues on this GitHub page.

Help is also available from the DENSITY | secr forum at phidot.org or the Google group secr.

See www.otago.ac.nz/density for general background.

secr's Issues

Support exact integral for detectfn (where possible)

While working on #4 , I've been looking into other ways of speeding up the computation.
One option would be to use the exact CDF instead of approximating the integral via Numer::Integrate where possible.

E.g.,the half normal function (zhhnrC) has an exact solution (found via WolframAlpha, this includes the y calculations).

This gives ~10x speedup, reducing computation time from 40 to ca. 4 minutes in my case. I've confirmed that results are the same. This is of course limited to convex polygons though.

Do you think this is worth adding? I'd be happy to contribute this if yes :)

secr.fit failed initial values not found

Hello,
I am trying to fit a secr model to some camera trap data and I've tried a few different inputs I found when trying to find a solution and they all yield the same error/warning :

library(secr)
CH <- read.capthist(captfile = "DEC21cap2.txt",trapfile = "DEc21trap.txt", skip=1, detector = "proximity",covnames = NULL,binary.usage=TRUE)
No errors found :-)
my_traps <- traps(CH)
my_mask <- make.mask(my_traps, buffer = 3500, spacing = 1200, type = "trapbuffer")
m0<-secr.fit(CH,detectfn = "HHN", mask = my_mask, model = list(D=~1, lambda0=~1, sigma=~1))
Checking data
Preparing detection design matrices
Preparing density design matrix
Finding initial parameter values...
Warning messages:
1: In autoini(capthist = ch, mask = msk, binomN = tempbinomN, adjustg0 = details$binomN[1] == :
'dbar' close to zero; using detector spacing instead
2: In makeStart(start, parindx, capthist, mask, detectfn, link, details, :
'secr.fit' failed because initial values not found (data sparse?); specify transformed values in 'start'

initialsigma <-RPSV(captdata, CC=TRUE)[[1]]
cat("Quick and biased estimate of sigma=", initialsigma, "m/n")
Quick and biased estimate of sigma= 25.62888 m/n
fit <- secr.fit(CH, buffer = 4*initialsigma, trace = FALSE)
Warning messages:
1: In autoini(capthist = ch, mask = msk, binomN = tempbinomN, adjustg0 = details$binomN[1] == :
'dbar' close to zero; using detector spacing instead
2: In makeStart(start, parindx, capthist, mask, detectfn, link, details, :
'secr.fit' failed because initial values not found (data sparse?); specify transformed values in 'start'

library(secr)

ftraps<-read.csv("DEC21trap.csv")
names(ftraps)[2:3]<-c('x','y')
qtraps<-read.traps(data = ftraps, detector = "proximity")
cap<-read.csv("DEC21cap2.csv")
Dcap<-cap[,c(1,2,3,4)]
CH<-make.capthist(Dcap,qtraps,fmt = "trapID",noccasions = 38)
summary(CH, terse = TRUE)
Occasions Detections Animals Detectors
38 38 22 103
mask<-make.mask(qtraps,spacing = 1200)
Cats.secr0<-secr.fit(CH,model = list(D1,g01,sigma~1),mask = mask)
Checking data
Preparing detection design matrices
Preparing density design matrix
Finding initial parameter values...
Warning messages:
1: In autoini(capthist = ch, mask = msk, binomN = tempbinomN, adjustg0 = details$binomN[1] == :
'dbar' close to zero; using detector spacing instead
2: In makeStart(start, parindx, capthist, mask, detectfn, link, details, :
'secr.fit' failed because initial values not found (data sparse?); specify transformed values in 'start'
DEC21trap.csv
DEC21cap2.csv

secr.fit does not allow start values from previous model with fixedbeta

secr.fit fails with start values from a previous model fitted with fixed beta values (details$fixedbeta)

see makeStart.R