c++ error: dpotrf A.alpha failed, when running spPGOcc

[agentilis]

Thank you so much for the excellent work. I was excited to see 0.3.0 come out with random intercepts in the occupancy formulas. I have run into an issue with model execution and am stuck with how to resolve it. Any advice would be appreciated.

I receive the error "c++ error: dpotrf A.alpha failed", when running spPGOcc. Sometimes it fails on chain 1, and sometimes on chain 2. I have n.report = 80. When it fails, it fails before the first update (at 25%).

spOccupancy 0.3.0, R 4.1.3, Windows 11, 64-bit, Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz, 32.0 GB RAM.

Dataset: 3360 survey points, up to 15 replicates, across 6 western states.

Settings n.batch = 320 # will increase later batch.length = 25 NNGP = TRUE n.neighbors = 8 n.burn = 2000 n.thin = 4 n.chains = 2 # will increase later

This runs fine: m._.bjns2 <- spPGOcc(occ.formula = ~ (1|year) + (1|stratum), det.formula = ~ broad + julian + noise + start + I(start^2), data = flowList, n.omp.threads=n.omp.threads,k.fold.threads=k.fold.threads, n.batch = n.batch, batch.length = batch.length, accept.rate = 0.43, cov.model = "exponential", NNGP = NNGP, n.neighbors = n.neighbors, n.burn = n.burn, n.thin = n.thin, n.chains = n.chains, verbose = T, n.report=n.report) #, , k.fold = k.fold)

This causes the error (adding "elev" to the occupancy formula): m.e <- spPGOcc(occ.formula = ~ (1|year) + (1|stratum) + elev, det.formula = broad + julian + noise + start + I(start^2), data = flowList, n.omp.threads=n.omp.threads,k.fold.threads=k.fold.threads, n.batch = n.batch, batch.length = batch.length, accept.rate = 0.43, cov.model = "exponential", NNGP = NNGP, n.neighbors = n.neighbors, n.burn = n.burn, n.thin = n.thin, n.chains = n.chains, verbose = T, n.report=n.report)#, k.fold = k.fold)

However, this runs fine (keeping "elev", but removing "I(start^2)" from detection model). m.e <- spPGOcc(occ.formula = ~ (1|year) + (1|stratum) + elev, det.formula = broad + julian + noise + start, data = flowList, n.omp.threads=n.omp.threads,k.fold.threads=k.fold.threads, n.batch = n.batch, batch.length = batch.length, accept.rate = 0.43, cov.model = "exponential", NNGP = NNGP, n.neighbors = n.neighbors, n.burn = n.burn, n.thin = n.thin, n.chains = n.chains, verbose = T, n.report=n.report)#, k.fold = k.fold)

Any suggestions?

I would be happy to share my dataset privately, or to help debug the issue.

Thank you for your time and effort.

[doserjef]

Hi Rob,

Thanks for reaching out with the error, and for providing all the really useful info. Sorry you're running into this issue. That error indicates that there is a problem when computing the detection regression coefficients in the MCMC, although it could be caused by a problem somewhere else in the code. All the code you have there looks fine, so would you be able to send me the data set and R script via email (doserjef@msu.edu) and I'll take a look more closely? My guess is there is something with your specific data set that spPGOcc isn't expecting when including that specific combination of covariates.

Thanks,

Jeff

[doserjef]

Error was occurring as a result of very large, non-standardized covariates in the model, which was resolved by standardizing the variables or scaling them to smaller values.