Hello and apologies if this isn’t the right place for this!
I’m using brms to deal with autocorrelation (both spatial and temporal) in an analysis of remote sensing data. The preliminary models with limited subsets of the data (n = ~200) work great, but I’m getting an error message at higher sample sizes (~ n > 1000, but its not consistent) :
Compiling Stan program...
Start sampling
Error in mod$fit_ptr() :
Exception: variable does not exist; processing stage=data initialization; variable name=eigenMsar; base type=vector_d (in 'model750d2c580e_5c0c99c82ea5eda165ec4d5a5340d18e' at line 61)
Here’s the model formula and call to brm:
brms::brmsformula(
respons ~ pnt_prd + tm_snc + SGU + z_bins_fac2 + fSCA + prcp + tmin + tmax +
sar(M = w0, type = 'error') +
ar(time = year, gr = pixel, p = 1),
family = gaussian()
brms::brm(
formula = f0, data = cb_df, data2 = list(w0 = w0),
prior = NULL, iter = 2000, cores = 4, chains = 4,
silent = 0
)
It works fine without the SAR term. I’ve tried tracking down the error but, at least on the R end, eigenMsar is always there when I go looking for it with the debugger.
Here’s the head and structure of the data:
pnt_prd tm_snc SGU z_bins_fac2 fSCA prcp tmin tmax year pixel x y
1 46.84986 0 SandyBottoms Very High 0.00128491 22.21589 2.9624949 18.21315 1999 1434291 -1323303 1824192
2 36.27044 0 SandyBottoms Very High 0.00000000 22.49830 2.8694968 18.11171 1999 1258425 -1322283 1827192
3 49.37712 0 SandyBottoms Very High 0.00000000 27.26958 2.4275658 17.06963 1999 1260171 -1322673 1827162
4 30.97303 0 SandyBottoms Very High 0.00000000 22.61864 2.8091314 18.05812 1999 1258436 -1321953 1827192
5 44.32429 0 SandyBottoms Very High 0.00000000 22.46512 2.8859658 18.12633 1999 1258422 -1322373 1827192
6 35.21775 0 SandyBottoms Very High 0.00000000 40.61501 -0.6463861 13.21699 1999 1261940 -1322373 1827132
'data.frame': 14806 obs. of 33 variables:
$ respons : num 45.7 37.5 49.6 29.3 44 ...
$ pnt_prd : num 46.8 36.3 49.4 31 44.3 ...
$ year : int 1999 1999 1999 1999 1999 1999 1999 1999 1999 1999 ...
$ pixel : int 1434291 1258425 1260171 1258436 1258422 1261940 1261939 1251401 1263696 1427252 ...
$ fSCA : num 0.00128 0 0 0 0 ...
$ tm_snc : num 0 0 0 0 0 0 0 0 0 0 ...
$ SGU : chr "SandyBottoms" "SandyBottoms" "SandyBottoms" "SandyBottoms" ...
$ cst_dst : num 715 849 977 762 857 ...
$ x : num -1323303 -1322283 -1322673 -1321953 -1322373 ...
$ y : num 1824192 1827192 1827162 1827192 1827192 ...
$ prcp : num 22.2 22.5 27.3 22.6 22.5 ...
$ tmax : num 18.2 18.1 17.1 18.1 18.1 ...
$ tmin : num 2.96 2.87 2.43 2.81 2.89 ...
$ z_bins_fac2: Factor w/ 4 levels "High","Low","Very High",..: 3 3 3 3 3 3 3 3 3 3 ...
And the code to generate the spatial weights, for context:
s0 <- sp::SpatialPoints(cb_df[, c('x', 'y')], proj4string = raster::crs("+proj=aea +lat_0=23 +lon_0=-96 +lat_1=29.5 +lat_2=45.5 +x_0=0 +y_0=0 +ellps=GRS80 +units=m +no_defs"))
d0 <- spdep::dnearneigh(s0, d1 = 0, d2 = 1000)
w0 <- spdep::nb2listwdist(d0, s0, 'idw', 'W', 1)
Attached package and R information:
Attached packages:
spdep_1.1-11 sf_1.0-3 spData_2.0.1 brms_2.16.1 Rcpp_1.0.7 raster_3.5-2 sp_1.4-5
R version 4.1.1 (2021-08-10)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 21.04
Using linux kernel 5.11.
A sample of the data in package data format is at: GitHub - bmcnellis/testBRMS: Debugging repo for BRMS under the data directory.
Any help would be much appreciated!