near_strings2 {ptools}

Strings of Near Repeats using KDtrees

Description

Identifies cases that are nearby each other in space/time

Usage

near_strings2(dat, id, x, y, tim, DistThresh, TimeThresh, k = 300, eps = 1e-04)

Arguments

dat

data frame

id

string for id variable in data frame (should be unique)

x

string for variable that has the x coordinates

y

string for variable that has the y coordinates

tim

string for variable that has the time stamp (should be numeric or datetime)

DistThresh

scaler for distance threshold (in whatever units x/y are in)

TimeThresh

scaler for time threshold (in whatever units tim is in)

k

the k for the max number of neighbors to grab in the nn2 function in RANN package

eps

the nn2 function returns <=, so to return less (like near_strings1()), needs a small fudge factor

Details

This function returns strings of cases nearby in space and time. Useful for near-repeat analysis, or to identify potentially duplicate cases. This particular function uses kdtrees (from the RANN library). For very large data frames, this will run quite a bit faster than near_strings1 (although still may run out of memory). And it is not 100% guaranteed to grab all of the pairs. Tests I have done on my machine ~100k rows takes around 2 minutes with this code.

Value

A data frame that contains the ids as row.names, and two columns:

• CompId, a unique identifier that lets you collapse original cases together

• CompNum, the number of linked cases inside of a component

References

Wheeler, A. P., Riddell, J. R., & Haberman, C. P. (2021). Breaking the chain: How arrests reduce the probability of near repeat crimes. Criminal Justice Review, 46(2), 236-258.

See Also

near_strings1(), which uses loops but is guaranteed to get all pairs of cases and should be memory safe.

Examples

# Simplified example showing two clusters
s <- c(0,0,0,4,4)
ccheck <- c(1,1,1,2,2)
dat <- data.frame(x=1:5,y=0,
                  ti=s,
                  id=1:5)
res1 <- near_strings2(dat,'id','x','y','ti',2,1)
print(res1)


# This runs faster than near_strings1
library(sp)
nyc_shoot$id <- 1:nrow(nyc_shoot)  #incident ID can have dups
print(Sys.time())
res <- near_strings2(nyc_shoot@data,id='id',x='X_COORD_CD',y='Y_COORD_CD',
                     tim='OCCUR_DATE',DistThresh=1500,TimeThresh=3)
print(Sys.time()) #around 4 seconds on my machine
head(res)