Intersection for lists of numeric data

Question

I know that Intersection is a pattern matching function and will not evaluate list elements to determine if two or more list have equal elements. So my question is, what can be done to do this? Say, determine if two lists have numerically equal elements within some tolerance dx?

Answer 1

Closely related questions:

• DeleteDuplicates[] does not work as expected on floating point values
• How to remove duplicates from set of machine precision 2D points?
• The conflated two uses of SameTest within Intersection

---

The most direct method is to make use of the SameTest option:

a = {1, 2.2, 3.14159};
b = {11/5, Pi, 1/2, 1/4};

Intersection[a, b, SameTest -> (Abs[# - #2] < 0.0001 &)]

{2.2, 3.14159}

This however requires a pairwise comparison that has a much higher computational complexity than the default method. A different method that does not require a pairwise compare will be much faster on long lists. Iff each list is internally free of duplicates and your tolerance requirement can be relaxed to rounding you could use this:

Cases[GatherBy[Join[b, a], Round[#, 0.0001] &], {x_, _} :> x]

{11/5, π}

Note here that the exact values are returned despite the use of Round, because list b is given first in Join.

The speed advantage can be great:

a = RandomReal[99, 1000];
b = RandomReal[99, 1000];

Intersection[a, b, SameTest -> (Abs[# - #2] < 0.0001 &)]       // Timing // First
Cases[GatherBy[Join[b, a], Round[#, 0.0001] &], {x_, __} :> x] // Timing // First

1.638

0.0011728

If a list is not internally free of duplicates any of those values will be returned as well:

a = {1, 2.2, 3.14159};
b = {11/5, Pi, 1/2, 0.5}; (* note 1/2 and 0.5 *)

Cases[GatherBy[Join[b, a], Round[#, 0.0001] &], {x_, __} :> x]

{11/5, π, 1/2}

Answer 2

Couldn't find a way to do it with built-in functions, but here's a manual loop that's not much slower than Mr.Wizard's solution, but has no problems with duplicates and rounding:

TolerantIntersection[a_List, b_List, dx_] := 
  Module[{sb = Sort@b, i = 1, imax = Length@b, dif},
    Select[Sort@a,
      (While[(dif = sb[[i]] - #) <= -dx, 
         If[i == imax, Return@False]; ++i];
      dif < dx) &]]

P.S. I edited to make the comparisons with dx strict. 'Not much slower' - I mean, in the same order of magnitude, compared to Intersection, actually it's 2-3 times slower. Note, that it's not equivalent to Intersection, in cases like

a = {0.00011, 0.00012};
b = {0.0002};

Intersection returns one arbitrary(?) element from a, my function returns both, which I believe is more correct.