NEARESTPOINT

NEARESTPOINT - find the nearest value in another vector

Description

NEARESTPOINT - find the nearest value in another vector
 
IND = NEARESTPOINT(X,Y) finds the value in Y which is the closest to each value in X, so that abs(Xi-Yk) => abs(Xi-Yj) when k is not equal to j.
IND contains the indices of each of these points.

Example:
  NEARESTPOINT([1 4 12],[0 3]) -> [1 2 2]

[IND,D] = ... also returns the absolute distances in D,that is D == abs(X - Y(IND))
NEARESTPOINT(X, Y, M) specifies the operation mode M:
1) 'nearest' : default, same as above
2) 'previous': find the points in Y that just precedes a point in X
NEARESTPOINT([1 4 12],[0 3],'previous') -> [1 1 1]
3) 'next' : find the points in Y that directly follow a point in X
NEARESTPOINT([1 4 12],[0 3],'next') -> [2 NaN NaN]
 
If there is no previous or next point in Y for a point X(i), IND(i)
will be NaN.
 
X and Y may be unsorted.
 
This function is quite fast, and especially suited for large arrays with
time data. For instance, X and Y may be the times of two separate events, like simple and complex spike data of a neurophysiological study.
Run NEARESTPOINT('test') to see a test of it's running time

• Acknowledgements: This submission has inspired the following:; Efficient K-Nearest Neighbor Search using JIT, IPDM: Inter-Point Distance Matrix
• MATLAB release: MATLAB 6.5 (R13)

[urs (us) schwarz]

nice code, jos, albeit very long
however, how does your engine compare with our pedestrian one (error checks not shown):
% some data
a=1.1*[1 4 12 -11];
b=1.2*[0 -3 10 10 nan];
% the engine
[ix,ix]=min(abs(repmat(b.',1,size(a,2))-...
repmat(a,size(b,2),1)),[],2);
% the result
a,[b;ix.']
as ever, just a thought
us

[Jos van der Geest]

Thanks us, for sharing your thoughts.
<nearestpoint> will be especially usefull for _large_ datasets, in which repmat (twice) will cause memory problems.
For small datasets, either for-loops, or repmat will produce similar timings.
did you run nearestpoint('test')?

[urs (us) schwarz]

yes, it is tuned to handle LARGE data sets as can be seen easily by slightly modifying the test suite: there are striking differences depending on the overall size of as well as the difference between the two vectors between this and the repmat approach
please note: line 114 crashes if(f) X is nx1 or 1xn and Y is 1x1 - someone ought to look into this... :-)
otherwise, a fine and useful piece of code
us

[John D'Errico]

Interesting submission. Careful coding and well
documented.

[David Wagener]

Thanks this did exactly what I was looking for!!

[Scott Burnside]

Nicely written. Quite fast. Thanks for the effort.

[Drew Saunders]

Perfect for my app!

[Eduardo Gonzalez]

Excellent, perfect for me.

[Nick Clark]

Nice, does the job

[Michael Ashby]

Simple to use and understand. Thanks for the code and helpful documentation.