[R] [External] Re: Selecting a minimum value of an attribute associated with point values neighboring a given point and assigning it as a new attribute
Eric Berger
er|cjberger @end|ng |rom gm@||@com
Tue Nov 8 07:52:10 CET 2022
## Some further comments and approaches, divided into 3 sections
##
## Section 1: Micha's modified code
## Section 2: Eric's modified code:
## a) uses Micha's dist_matrix from dnearneigh instead of the f() function
## b) add check that it gets the same solution as Micha's modified code
## Section 3: Solution explicitly using a graph (via package igraph)
## ###########################################################
## Section 1: Micha's modified code
##
## Major change: OP wanted the Min Conc value
## to include the Conc of the point itself
## Minor Changes: no need to write/read csv file;
## change MAX_DIST to 50 for fewer neighbors
library(sf)
library(spdep)
## Prepare fictitious data
## Create a data.frame with n random points
set.seed(234) ## for reproducibility
N <- 1000L
LON <- runif(N, -70.0, -69.0)
LAT <- runif(N, 42.0, 43.0)
Conc <- runif(N, 90000, 100000)
df <- data.frame(LON, LAT, Conc)
## Create a distance matrix for all points,
## which contains indices to those points within
## a certain buffer distance, just as you did in your example.
MAX_DIST <- 50
pts <- st_as_sf(df, coords=c('LON', 'LAT'), crs=4326)
dist_matrix <- dnearneigh(pts, 0, MAX_DIST, use_s2=TRUE)
cat("Average number of neighbors within cutoff = ",
mean(unlist(lapply(dist_matrix,length))),"\n")
## Function to get minimum Conc values for one row of distance matrix
MinConc <- function(x, lst, pts) {
## x is an index to a single point,
## lst is a list of point indices from distance matrix
## that are within the buffer distance
Concs <- lapply(lst, function(p) {
pts$Conc[p]
})
## return(min(c(Concs[[1]])) - ## original code - forgets to
include the point itself
return(min(c(Concs[[1]], pts$Conc[x]))) ## modified code
}
## Now apply this function to all points in pts
Conc_min <- lapply(1:N, function(i) {
MinConc(i, dist_matrix[i], pts)
})
Conc_min <- data.frame("Conc_min" = as.numeric(Conc_min))
# Add back as new attrib to original points sf object
pts_with_min <- do.call(cbind, c(pts, Conc_min))
## ###########################################################
## Section 2: Eric's modified code:
A <- matrix(0,N,N)
z <- sapply(1:N, \(i) A[i, dist_matrix[[i]]] <<- 1) ## z is ignored
B <- A + diag(N)
C <- diag(Conc)
D <- B %*% C
D[D==0] <- NA
Conc_min.eric <- apply(D,MAR=1,\(v) min(v,na.rm=TRUE) )
test <- identical(Conc_min$Conc_min, Conc_min.eric)
cat("test = ", test, "\n")
## ###########################################################
## Section 3: Solution explicitly using a graph (via package igraph)
## For those with some familiarity with graphs, the matrix 'B' is
## an adjacency matrix. This suggests using graphs explicitly to solve
## the problem. Here is how to rewrite my code using graphs.
library(igraph)
g <- graph_from_adjacency_matrix(B,"undirected")
g <- set_vertex_attr(g, "Conc", 1:N, Conc )
Conc_min.igraph <- sapply(1:N, \(i) min(vertex_attr(g,"Conc",neighbors(g,i))))
test.igraph <- identical(Conc_min$Conc_min, Conc_min.igraph)
cat("test.igraph = ", test.igraph, "\n")
Eric
On Mon, Nov 7, 2022 at 3:11 PM Micha Silver <tsvibar using gmail.com> wrote:
>
> Eric's solution notwithstanding, here's a more "spatial" approach.
>
>
> I first create a fictitious set of 1000 points (and save to CSV to
> replicate your workflow)
>
> library(sf)
> library(spdep)
>
> # Prepare fictitious data
> # Create a data.frame with 1000 random points, and save to CSV
> LON <- runif(1000, -70.0, -69.0)
> LAT <- runif(1000, 42.0, 43.0)
> Conc <- runif(1000, 90000, 100000)
> df <- data.frame(LON, LAT, Conc)
> csv_file = "/tmp/pts_testdata.csv"
> write.csv(df, csv_file)
>
>
> Now read that CSV back in directly as an sf object (No need for the old
> SpatialPointsDataFrame). THen create a distance matrix for all points,
> which contains indicies to those points within a certain buffer
> distance, just as you did in your example.
>
>
> # Read back in as sf object, including row index
> pts <- st_as_sf(read.csv(csv_file), coords=c('LON', 'LAT'), crs=4326)
> dist_matrix <- dnearneigh(pts, 0, 100, use_s2=TRUE) # use_s2 since these
> are lon/lat
>
> Now I prepare a function to get the minimum Conv value among all points
> within the buffer distance to a given single point:
> # Function to get minimum Conc values for one row of distance matrix
> MinConc <- function(x, lst, pts) {
> # x is an index to a single point,
> # lst is a list of point indices from distance matrix
> # that are within the buffer distance
> Concs <- lapply(lst, function(p) {
> pts$Conc[p]
> })
> return(min(Concs[[1]]))
> }
>
> Next run that function on all points to get a list of minimum Conv
> values for all points, and merge back to pts.
>
>
> # Now apply this function to all points in pts
> Conc_min <- lapply(pts$X, function(i){
> MinConc(i, dist_matrix[i], pts)
> })
> Conc_min <- data.frame("Conc_min" = as.integer(Conc_min))
>
> # Add back as new attrib to original points sf object
> pts_with_min <- do.call(cbind, c(pts, Conc_min))
>
> HTH,
>
> Micha
>
>
>
> On 06/11/2022 18:40, Duhl, Tiffany R. wrote:
> > Thanks so much Eric!
> >
> > I'm going to play around with your toy code (pun intended) & see if I can make it work for my application.
> >
> > Cheers,
> > -Tiffany
> > ________________________________
> > From: Eric Berger <ericjberger using gmail.com>
> > Sent: Sunday, November 6, 2022 10:27 AM
> > To: Bert Gunter <bgunter.4567 using gmail.com>
> > Cc: Duhl, Tiffany R. <Tiffany.Duhl using tufts.edu>; R-help <R-help using r-project.org>
> > Subject: [External] Re: [R] Selecting a minimum value of an attribute associated with point values neighboring a given point and assigning it as a new attribute
> >
> > Whoops ... left out a line in Part 2. Resending with the correction
> >
> > ## PART 2: You can use this code on the real data with f() defined appropriately
> > A <- matrix(0,N,N)
> > v <- 1:N
> > ## get the indices (j,k) where j < k (as columns in a data.frame)
> > idx <- expand.grid(v,v) |> rename(j=Var1,k=Var2) |> filter(j < k)
> > u <- sapply(1:nrow(idx),
> > \(i){ j <- idx$j[i]; k <- idx$k[i]; A[j,k] <<- f(j,k,myData) })
> > B <- A + t(A) + diag(N)
> > C <- diag(myData$Conc)
> > D <- B %*% C
> > D[D==0] <- NA
> > myData$Conc_min <- apply(D,MAR=1,\(v){min(v,na.rm=TRUE)})
> > print(head(myData))
> >
> > On Sun, Nov 6, 2022 at 5:19 PM Eric Berger <ericjberger using gmail.com> wrote:
> >> Hi Tiffany,
> >> Here is some code that might help with your problem. I solve a "toy"
> >> problem that is conceptually the same.
> >> Part 1 sets up my toy problem. You would have to replace Part 1 with
> >> your real case. The main point is to define
> >> a function f(i, j, data) which returns 0 or 1 depending on whether the
> >> observations in rows i and j in your dataset 'data'
> >> are within your cutoff distance (i.e. 50m).
> >>
> >> You can then use Part 2 almost without changes (except for changing
> >> 'myData' to the correct name of your data).
> >>
> >> I hope this helps,
> >> Eric
> >>
> >> library(dplyr)
> >>
> >> ## PART 1: create fake data for minimal example
> >> set.seed(123) ## for reproducibility
> >> N <- 5 ## replace by number of locations (approx 9000 in your case)
> >> MAX_DISTANCE <- 2 ## 50 in your case
> >> myData <- data.frame(x=rnorm(N),y=rnorm(N),Conc=sample(1:N,N))
> >>
> >> ## The function which you must re-define for your actual case.
> >> f <- function(i,j,a) {
> >> dist <- sqrt(sum((a[i,1:2] - a[j,1:2])^2)) ## Euclidean distance
> >> as.integer(dist < MAX_DISTANCE)
> >> }
> >>
> >> ## PART 2: You can use this code on the real data with f() defined appropriately
> >> A <- matrix(0,N,N)
> >> ## get the indices (j,k) where j < k (as columns in a data.frame)
> >> idx <- expand.grid(v,v) |> rename(j=Var1,k=Var2) |> filter(j < k)
> >> u <- sapply(1:nrow(idx),\(i){ j <- idx$j[i]; k <- idx$k[i]; A[j,k] <<-
> >> f(j,k,myData) })
> >> B <- A + t(A) + diag(N)
> >> C <- diag(myData$Conc)
> >> D <- B %*% C
> >> D[D==0] <- NA
> >> myData$Conc_min <- apply(D,MAR=1,\(v){min(v,na.rm=TRUE)})
> >> print(head(myData))
> >>
> >>
> >> On Sat, Nov 5, 2022 at 5:14 PM Bert Gunter <bgunter.4567 using gmail.com> wrote:
> >>> Probably better posted on R-sig-geo.
> >>>
> >>> -- Bert
> >>>
> >>> On Sat, Nov 5, 2022 at 12:36 AM Duhl, Tiffany R. <Tiffany.Duhl using tufts.edu>
> >>> wrote:
> >>>
> >>>> Hello,
> >>>>
> >>>> I have sets of spatial points with LAT, LON coords (unprojected, WGS84
> >>>> datum) and several value attributes associated with each point, from
> >>>> numerous csv files (with an average of 6,000-9,000 points in each file) as
> >>>> shown in the following example:
> >>>>
> >>>> data<- read.csv("R_find_pts_testdata.csv")
> >>>>
> >>>>> data
> >>>> ID Date Time LAT LON Conc
> >>>> Leg.Speed CO2 H2O BC61 Hr Min Sec
> >>>> 1 76 4/19/2021 21:25:38 42.40066 -70.98802 99300 0.0 mph 428.39 9.57
> >>>> 578 21 25 38
> >>>> 2 77 4/19/2021 21:25:39 42.40066 -70.98802 96730 0.0 mph 428.04 9.57
> >>>> 617 21 25 39
> >>>> 3 79 4/19/2021 21:25:41 42.40066 -70.98802 98800 0.2 mph 427.10 9.57
> >>>> 1027 21 25 41
> >>>> 4 80 4/19/2021 21:25:42 42.40066 -70.98802 96510 2 mph 427.99 9.58
> >>>> 1381 21 25 42
> >>>> 5 81 4/19/2021 21:25:43 42.40067 -70.98801 95540 3 mph 427.99 9.58
> >>>> 1271 21 25 43
> >>>> 6 82 4/19/2021 21:25:44 42.40068 -70.98799 94720 4 mph 427.20 9.57
> >>>> 910 21 25 44
> >>>> 7 83 4/19/2021 21:25:45 42.40069 -70.98797 94040 5 mph 427.18 9.57
> >>>> 652 21 25 45
> >>>> 8 84 4/19/2021 21:25:46 42.40072 -70.98795 95710 7 mph 427.07 9.57
> >>>> 943 21 25 46
> >>>> 9 85 4/19/2021 21:25:47 42.40074 -70.98792 96200 8 mph 427.44 9.56
> >>>> 650 21 25 47
> >>>> 10 86 4/19/2021 21:25:48 42.40078 -70.98789 93750 10 mph 428.76 9.57
> >>>> 761 21 25 48
> >>>> 11 87 4/19/2021 21:25:49 42.40081 -70.98785 93360 11 mph 429.25 9.56
> >>>> 1158 21 25 49
> >>>> 12 88 4/19/2021 21:25:50 42.40084 -70.98781 94340 12 mph 429.56 9.57
> >>>> 107 21 25 50
> >>>> 13 89 4/19/2021 21:25:51 42.40087 -70.98775 92780 12 mph 428.62 9.56
> >>>> 720 21 25 51
> >>>>
> >>>>
> >>>> What I want to do is, for each point, identify all points within 50m of
> >>>> that point, find the minimum value of the "Conc" attribute of each nearby
> >>>> set of points (including the original point) and then create a new variable
> >>>> ("Conc_min") and assign this minimum value to a new variable added to
> >>>> "data".
> >>>>
> >>>> So far, I have the following code:
> >>>>
> >>>> library(spdep)
> >>>> library(sf)
> >>>>
> >>>> setwd("C:\\mydirectory\\")
> >>>> data<- read.csv("R_find_pts_testdata.csv")
> >>>>
> >>>> #make sure the data is a data frame
> >>>> pts <- data.frame(data)
> >>>>
> >>>> #create spatial data frame and define projection
> >>>> pts_coords <- cbind(pts$LON, pts$LAT)
> >>>> data_pts <- SpatialPointsDataFrame(coords= pts_coords,
> >>>> data=pts, proj4string = CRS("+proj=longlat +datum=WGS84"))
> >>>>
> >>>> #Re-project to WGS 84 / UTM zone 18N, so the analysis is in units of m
> >>>> ptsUTM <- sf::st_as_sf(data_pts, coords = c("LAT", "LON"), remove = F)%>%
> >>>> st_transform(32618)
> >>>>
> >>>> #create 50 m buffer around each point then intersect with points and
> >>>> finally find neighbors within the buffers
> >>>> pts_buf <- sf::st_buffer(ptsUTM, 50)
> >>>> coords <- sf::st_coordinates(ptsUTM)
> >>>> int <- sf::st_intersects(pts_buf, ptsUTM)
> >>>> x <- spdep::dnearneigh(coords, 0, 50)
> >>>>
> >>>> Now at this point, I'm not sure what to either the "int" (a sgbp list) or
> >>>> "x" (nb object) objects (or even if I need them both)
> >>>>
> >>>>> int
> >>>> Sparse geometry binary predicate list of length 974, where the predicate
> >>>> was `intersects'
> >>>> first 10 elements:
> >>>> 1: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 2: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 3: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 4: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 5: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 6: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 7: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 8: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>> 9: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, ...
> >>>>
> >>>>> x
> >>>> Neighbour list object:
> >>>> Number of regions: 974
> >>>> Number of nonzero links: 34802
> >>>> Percentage nonzero weights: 3.668481
> >>>> Average number of links: 35.73101
> >>>>
> >>>> One thought is that maybe I don't need the dnearneigh function and can
> >>>> instead convert "int" into a dataframe and somehow merge or associate
> >>>> (perhaps with an inner join) the ID fields of the buffered and intersecting
> >>>> points and then compute the minimum value of "Conc" grouping by ID:
> >>>>
> >>>>> as.data.frame(int)
> >>>> row.id col.id
> >>>> 1 1 1
> >>>> 2 1 2
> >>>> 3 1 3
> >>>> 4 1 4
> >>>> 5 1 5
> >>>> 6 1 6
> >>>> 7 1 7
> >>>> 8 1 8
> >>>> 9 1 9
> >>>> 10 1 10
> >>>> 11 1 11
> >>>> 12 1 12
> >>>> 13 1 13
> >>>> 14 1 14
> >>>> 15 1 15
> >>>> 16 1 16
> >>>> 17 1 17
> >>>> 18 1 18
> >>>> 19 2 1
> >>>> 20 2 2
> >>>> 21 2 3
> >>>> 22 2 4
> >>>> 23 2 5
> >>>> 24 2 6
> >>>> 25 2 7
> >>>> 26 2 8
> >>>> 27 2 9
> >>>> 28 2 10
> >>>>
> >>>>
> >>>> So in the above example I'd like to take the minimum of "Conc" among the
> >>>> col.id points grouped with row.id 1 (i.e., col.ids 1-18) and assign the
> >>>> minimum value of this group as a new variable in data (Data$Conc_min), and
> >>>> do the same for row.id 2 and all the rest of the rows.
> >>>>
> >>>> I'm just not sure how to do this and I appreciate any help folks might
> >>>> have on this matter!
> >>>>
> >>>> Many thanks,
> >>>> -Tiffany
> >>>>
> >>>> [[alternative HTML version deleted]]
> >>>>
> >>>> ______________________________________________
> >>>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> >>>> https://stat.ethz.ch/mailman/listinfo/r-help
> >>>> PLEASE do read the posting guide
> >>>> http://www.R-project.org/posting-guide.html
> >>>> and provide commented, minimal, self-contained, reproducible code.
> >>>>
> >>> [[alternative HTML version deleted]]
> >>>
> >>> ______________________________________________
> >>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> >>> https://stat.ethz.ch/mailman/listinfo/r-help
> >>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> >>> and provide commented, minimal, self-contained, reproducible code.
> > Caution: This message originated from outside of the Tufts University organization. Please exercise caution when clicking links or opening attachments. When in doubt, email the TTS Service Desk at it using tufts.edu<mailto:it using tufts.edu> or call them directly at 617-627-3376.
> >
> >
> > [[alternative HTML version deleted]]
> >
> > ______________________________________________
> > R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> > and provide commented, minimal, self-contained, reproducible code.
>
> --
> Micha Silver
> Ben Gurion Univ.
> Sde Boker, Remote Sensing Lab
> cell: +972-523-665918
>
More information about the R-help
mailing list