feat: support theta and tuple a-not-b

[ariesdevil]

based on #152 #153 #154

[tisonkun]

#154 is merged now.

I wonder if we'd have a tuple_a_not_b_test.rs integration test. And some common test code may be de-duplicate later.

[tisonkun]

I have some ideas on factoring out this test utility support module. Will do once this PR gets merged.

[tisonkun]

... to avoid too much #[cfg(feature = "...")] and #[allow(dead_code)].

[tisonkun]

a_not_b comes from datasketches-cpp and datasketches-java?

@leerho @proost what if we name this set operation as difference or exception, since a_not_b is somewhat strange to me. Or is it actually an idiom in this domain?

[tisonkun]

Based on the result, it looks like the same as set difference.

@ariesdevil What do you think?

[leerho]

a_not_b is a set difference. However the name a_not_b is a mnemonic that tells you the direction of the difference.

In the math of boolean logic one would write

A \ B := A AND NOT B = $A \cup !B$
or
B \ A := B AND NOT A = $B \cup !A$

Where "\" is the difference operator.
Note that A \ B != B \ A !

Although C++ allows user defined operators most languages do not. To call the difference function, we need to indicate the direction. We chose to do it by the position of the arguments, and the name of the function is a mnemonic of the direction.

Sketch result a_not_b(Sketch A, Sketch B);

[tisonkun]

@leerho Thanks for your information!

Yes the Rust impl also defines the API as:

let update_policy = DefaultUpdatePolicy::<u64>::default();
let mut a = TupleSketchBuilder::new(update_policy).build();
a.update("apple", 1);
a.update("banana", 1);

let mut b = TupleSketchBuilder::new(update_policy).build();
b.update("banana", 1);

let a_not_b = TupleAnotB::default();
let result = a_not_b.compute(&a, &b, true).unwrap();

So I generally agree that a_not_b provides more information and avoid the confusion of op direction.

Even if we can add methods or operator overrides to ThetaSketch/TupleSketch, it is not symmetry to the union and intersection interface.

One minor comment: any typical reason to name it TupleAnotB rather than TupleANotB? Weird CamelCase at first glance.

[tisonkun]

Generally LGTM.

Anything keeps this PR a draft before it's ready for review and merge?

[tisonkun]

Minor comments can be handled later.