Question

I have the following code with uses the FNN Package:

k <- knn(train, test, labels, k = 10, algorithm="cover_tree")

I am able to look at the data returned by the call to knn using the code below:

> attr(k, "nn.dist")[1,]
 [1] 1385.398 1687.901 1722.884 1846.694 1978.250 1998.157 2003.518 2004.494 2054.382 2059.128
> attr(k, "nn.index")[1,]
 [1] 983 362 170 303 914 843 480 489 474 355

What I would like to do is modify the values that are in nn.dist based on the values that are in nn.index.

I know this probably doesn’t make sense, but for the sake of argument, how would I go about setting each value in nn.dist to nn.dist * nn.index/1000?

So attr(k, "nn.dist")[1,1] would go from 1385.398 to 1385.398*983/1000 (1361.846)

and
attr(k, "nn.dist")[1,2] would go from 1687.901 to 1687.901*362/1000 (611.02) etc.

I am able to do this one at a time with the following code:

attr(k, "nn.dist")[1,1] = attr(k, "nn.dist")[1,1]*attr(k, "nn.index")[1,1]/1000

But I need a way to do this to every element, without writing it every time…

Answer 1

The basic operators *, +, / and - are vectorised and will work elementwise on an array.

Therefore it is as easy as

attr(k, "nn.dist") * attr(k, "nn.index") /1000

If you only want to do this to the first row of nn.dist, then

attr(k, "nn.dist")[1,]* attr(k, "nn.index")[1,] /1000

You could reassign to the nn.dist attribute of k, but I would suggest copying to ensure that you know that it is not the original output eg

k_adj <- k
attr(k_adj, "nn.dist") <- attr(k_adj, "nn.dist") * attr(k_adj, "nn.index") /1000

or you could simply add a new attribute to the original

 attr(k, 'nn.dist.adj') <- attr(k, "nn.dist") * attr(k, "nn.index") /1000