R help - May 2020 - gdistance::accCost fails when nrows is bigger than 90

Hello all,

It has been only 2 days since I started using R and I was trying movecost
library. Before explaining the unexpected behavior I should say that I?m calling
R function from Python using rpy2 package. That shouldn?t be an issue because
everything works as expected when a matrix of size 100x90 is given.

Okay now, basically I?m converting an 2D matrix to RasterLayer with appropriate
CRS and extent properties then transfer it to R runtime. A modified version of
Tobler?s hiking function is applied then a shortest-path is calculated. This
procedure works for a matrix with column size anywhere between 1 and memory?s
limit. However if row size of the matrix exceeds 90 then function is halted on
gdistance::accCost saying that:

"At structural_properties.c:5313 : cannot run Bellman-Ford algorithm,
Negative loop detected while calculating shortest paths?

It?s very interesting issue because I have explicitly tried ?88, 89, 90, 91 and
it halted on +90

I?ve created a gist in case you want to try the application on your computer.
Also I?ve included a clean version of the R script I was using bellow.

Thank you.

https://gist.github.com/DenizUgur/01e18edd03321f3d3928c3afb2bd7145
<https://gist.github.com/DenizUgur/01e18edd03321f3d3928c3afb2bd7145>

movecost <- function (dtm, origin, destin, time="s") {

  altDiff <- function(x){x[2] - x[1]}
  hd <- gdistance::transition(dtm, altDiff, 8, symm=FALSE)

  slope <- gdistance::geoCorrection(hd)

  cost_function <- function(x){ ifelse(x[adj] > 0, 1.3 * exp(-3.5 *
abs(x[adj] + 0.05)), 5.3 * exp(-3.5 * abs(x[adj] + 0.05))) }

  adj <- raster::adjacent(dtm, cells=1:ncell(dtm), pairs=TRUE, directions=8)
  speed <- slope
  speed[adj] <- cost_function(slope)
  speed <- speed * 0.278
  Conductance <- gdistance::geoCorrection(speed)

  accum_final <- gdistance::accCost(Conductance, sp::coordinates(origin))

  accum_final <- raster::mask(accum_final, dtm)

  sPath <- gdistance::shortestPath(Conductance, sp::coordinates(origin),
sp::coordinates(destin), output="SpatialLines")
  sPath$length <- rgeos::gLength(sPath, byid=TRUE)
  destin$cost <- raster::extract(accum_final, destin)
  
  results <- list("accumulated.cost.raster"=accum_final,
                  "LCPs"=sPath,
                  "dest.loc.w.cost"=destin)
}
	[[alternative HTML version deleted]]

Referring to my mail bellow, this situation is not a expected behavior. Do
you think that this is because of a fault in my codes or some bug in
gdistance? What could be the problem?

Thank you for your consideration.

---------- Forwarded message ---------
From: Deniz U?ur <deniz343 at gmail.com>
Date: Thu, May 14, 2020, 02:18
Subject: gdistance::accCost fails when nrows is bigger than 90
To: <r-help at r-project.org>


Hello all,

It has been only 2 days since I started using R and I was trying movecost
library. Before explaining the unexpected behavior I should say that I?m
calling R function from Python using rpy2 package. That shouldn?t be an
issue because everything works as expected when a matrix of size 100x90 is
given.

Okay now, basically I?m converting an 2D matrix to RasterLayer with
appropriate CRS and extent properties then transfer it to R runtime. A
modified version of Tobler?s hiking function is applied then a
shortest-path is calculated. This procedure works for a matrix with column
size anywhere between 1 and memory?s limit. However if row size of the
matrix exceeds 90 then function is halted on gdistance::accCost saying that:

"At structural_properties.c:5313 : cannot run Bellman-Ford algorithm,
Negative loop detected while calculating shortest paths?

It?s very interesting issue because I have explicitly tried ?88, 89, 90, 91
and it halted on +90

I?ve created a gist in case you want to try the application on your
computer. Also I?ve included a clean version of the R script I was using
bellow.

Thank you.

https://gist.github.com/DenizUgur/01e18edd03321f3d3928c3afb2bd7145

movecost <- function (dtm, origin, destin, time="s") {

altDiff <- function(x){x[2] - x[1]}
hd <- gdistance::transition(dtm, altDiff, 8, symm=FALSE)

slope <- gdistance::geoCorrection(hd)

cost_function <- function(x){ ifelse(x[adj] > 0, 1.3 * exp(-3.5 *
abs(x[adj]
+ 0.05)), 5.3 * exp(-3.5 * abs(x[adj] + 0.05))) }

adj <- raster::adjacent(dtm, cells=1:ncell(dtm), pairs=TRUE, directions=8)
speed <- slope
speed[adj] <- cost_function(slope)
speed <- speed * 0.278
Conductance <- gdistance::geoCorrection(speed)

accum_final <- gdistance::accCost(Conductance, sp::coordinates(origin))

accum_final <- raster::mask(accum_final, dtm)

sPath <- gdistance::shortestPath(Conductance, sp::coordinates(origin), sp::
coordinates(destin), output="SpatialLines")
sPath$length <- rgeos::gLength(sPath, byid=TRUE)
destin$cost <- raster::extract(accum_final, destin)
results <- list("accumulated.cost.raster"=accum_final,
"LCPs"=sPath,
"dest.loc.w.cost"=destin)
}

	[[alternative HTML version deleted]]