0

would really appreciate any help of rectifying my issue in R.

I have written a KDE function in R, and I am using MASS::kde2d as my reference, to plot and model crime data. The density values I am producing seem to be in the correct range, as well as the overall distribution of densities, but it appears to be flipped by 45 degrees through the plots origin.

Notice in the plot below that the shape is almost identical, but flipped.

Right plot: kde2d function. Left plot: my Kernel2D function

Any help would be greatly appreciated! I have attached my 2d function below if anyone can spot an error. Thank you!

The input of the function is a spatial lattice, with coordinates at the center of the grid cells, crime locations in latitude then longitude order, bandwidth h and a switch parameter. I am looking at including other kernels hence the inclusion. 

Kernel2D <- function(lattice,  crime.locations, h, kernel){
  if(is.data.frame(spatial.lattice)==FALSE) {lattice <- data.frame(lattice)}

  d <- fields::rdist(lattice, crime.locations)

  switch(kernel,


         'normal'={
           k=1
           cat('k = 1, then rescaled so sum of densities = N. ')
           cat("Normal - Function extends to boundary in all directions, applied to every location in the region regardless of h")

## This is the density calculation which iterates through the rows of d. I.e. for 
## each cell in the lattice, it does a calculation on that row where the row contains the
## distances from that cell to each crime location. (Not trying to patronize anyone here 
## with having dumb explanations! Trying to make it simple so I may understand it myself 
##  when explaining it to others haha) .
## I have assumed that this is correct as it it producing the same density shape as MASS::kde2d. 
## There is an error when associating the density values to the right grid. Hmm. Ill have more of a think now. 

           density <- apply(d, 1, function(x) sum((1/(2*pi*h**2))*exp(-(x**2)/(2*h**2))))
           k = nrow(crime.locations)/sum(density)
           density <- k*density},

         stop('Kernel Undefined. Available: triangular, uniform, negexp, normal, quartic'))

  return(density)}
Machavity
  • 30,841
  • 27
  • 92
  • 100
Lana Francis
  • 1
  • 2

1 Answers1

0

I think I may have found a reason why this is happening.

So there is clearly some error with associating density values with the correct grid when I am plotting.

So in the function I may be calculating over a lattice row by row, which returns a one column vector (in row by row order). I am potentially attaching this vector to a data frame (for example) but the rows of cells in the df are not in row by row order in the lattice, they are actually column by column. Thus when I set up the data for plotting, I am attaching densities to the wrong grids. And if the row vs column theory is correct, this would produce a plot that will be flipped.

Ah ha, this may be the issue. i will change my function to attach densities to the grids as its calculating so the right grid gets the correct density. Will get back to you with an update when I have tried this!!

Lana Francis
  • 1
  • 2