Set.insert vs array << x unless array.include?(x)

[thomthom]

I was curious of the difference between Set.insert vs array << x unless array.include?(x) when you want a collection with a unique set of values.

So I did some tests:

Test 1
A 10,000,000 times I tried to insert a number between 0-10.

t=Time.now;a=[];10000000.times{r=rand(10);a<<r unless a.include?(r)};puts Time.now - t
Result: 12.297

t=Time.now;a=Set.new;10000000.times{a.insert(rand(10))};puts Time.now - t
Result: 15.719

The array was faster here by three seconds. But that's with ten million iterations!


Test 2
But, when your collection of unique values increases:
t=Time.now;a=[];100000.times{r=rand(10000);a<<r unless a.include?(r)};puts Time.now - t
Result: 45.129

t=Time.now;a=Set.new;100000.times{a.insert(rand(10000))};puts Time.now - t
Result: 0.391

When your collection include a much higher number of elements - then the Array quickly starts to crumble. In this case, when the possible unique set was up to 100,000 the difference was 44.738 seconds!!!

I tried this in some of my scrips which on occasions require a unique set of entites (like collecting all vertices) and the speed difference was immense on larger models.

It doesn't take too many number of unique elements before Set becomes faster then array.include?.


Scaling
Just to illustrate how they scale:
Array.include?
t=Time.now;a=[];1000.times{r=rand(100);a<<r unless a.include?(r)};puts Time.now - t
0.0
t=Time.now;a=[];10000.times{r=rand(1000);a<<r unless a.include?(r)};puts Time.now - t
0.485
t=Time.now;a=[];100000.times{r=rand(10000);a<<r unless a.include?(r)};puts Time.now - t
42.906
t=Time.now;a=[];1000000.times{r=rand(100000);a<<r unless a.include?(r)};puts Time.now - t
Took too long. I guess it'd take ~4000.0s


Set.insert
t=Time.now;a=Set.new;1000.times{a.insert(rand(100))};puts Time.now - t
0.0
t=Time.now;a=Set.new;10000.times{a.insert(rand(1000))};puts Time.now - t
0.047
t=Time.now;a=Set.new;100000.times{a.insert(rand(10000))};puts Time.now - t
0.422
t=Time.now;a=Set.new;1000000.times{a.insert(rand(100000))};puts Time.now - t
4.937


Conclusion (not!)
Use the Set class when you require a unique set of values - unless your set of unique values is very small.
Set class is the rule - Array.include? is the exception.

Update
None of the above is the fastest. A better way has appeared. Collect everything into an array and apply .unique! at the end.
viewtopic.php?f=180&t=23760&p=222231#p222221

[thomthom]

Also, Set.include? is faster than Array.include?

[Fredo6]

Tom,

very useful actually.
It really looks that the processsing in C is a lot faster than in Ruby.

Thanks

Fredo

[thomthom]

Tom,

very useful actually.
It really looks that the processsing in C is a lot faster than in Ruby.

Thanks

Fredo

Set class makes more processing in C?
What I'm wondering is, the Set class in SU Ruby is not the same Set class you get in Ruby 1.8.0. The Set class in Ruby 1.8.0 includes Enumerable and the set.rb looks to be a pure ruby class that mixes arrays and hashes. From my Ruby 1.8.0. installation, the Set class seem to be a pure Ruby implementation. But maybe the Set class that comes with SU is a C implementation...? Maybe that's why they replaced it?

[thomthom]

I thought the speed difference was that the Set class used a more efficient Hash lookup as oppose to the Array functions that has to iterate the array every time... but that's just guesswork...

[Fredo6]

I thought the speed difference was that the Set class used a more efficient Hash lookup as oppose to the Array functions that has to iterate the array every time... but that's just guesswork...

You're right.
I personally use Hash whenever I want to store lists that have unique elements.
The list itself is obtained via Hash.values.

Maybe, as you are on this, you extend your benchmark to Hash and see how it compares with Set.

Fred

[Chris Fullmer]

Great info Thom! I had played around with sets and arrays before and I was not impressed with sets. But I did not experiment with amount of unique objects, which apparently has an adverse effect on arrays. Thanks,

Chris

[thomthom]

[quote="Fredo6"
Maybe, as you are on this, you extend your benchmark to Hash and see how it compares with Set.[/quote]
Will have a look at that.

[Nazz78]

How about using Array.uniq! method:

Test 1
t=Time.now;a=[];10000000.times{r=rand(10);a<<r unless a.include?(r)};puts Time.now - t
Result: 12.297
t=Time.now;a=Set.new;10000000.times{a.insert(rand(10))};puts Time.now - t
Result: 15.719
t=Time.now;a=[];10000000.times{r=rand(10);a<<r};a.uniq!; puts Time.now - t
Result: 7.753

Test 2
t=Time.now;a=[];100000.times{r=rand(10000);a<<r unless a.include?(r)};puts Time.now-t
Result: 40.97
t=Time.now;a=Set.new;100000.times{a.insert(rand(10000))};puts Time.now-t
Result: 0.377
t=Time.now;a=[];100000.times{r=rand(10000);a<<r};a.uniq!;puts Time.now-t
Result: 0.087

[thomthom]

I'll be damned!
Very interesting Jernej.

...looks like I need to do some more testing of my script and possibly refactor again.

So while the Array.include? is dead slow - the overhead of hash look-up is still faster than just adding everything into one big pile and so a single filtering afterwards...

[thomthom]

looking at the .uniq! source code: http://ruby-doc.org/core/classes/Array.src/M002215.html

Code: Select all

/*
*  call-seq:
*     array.uniq! -> array or nil
* 
*  Removes duplicate elements from _self_.
*  Returns <code>nil</code> if no changes are made (that is, no
*  duplicates are found).
*     
*     a = [ "a", "a", "b", "b", "c" ]
*     a.uniq!   #=> ["a", "b", "c"]
*     b = [ "a", "b", "c" ]
*     b.uniq!   #=> nil
*/

static VALUE
rb_ary_uniq_bang(ary)
    VALUE ary;
{
    VALUE hash, v, vv;
    long i, j;

    hash = ary_make_hash(ary, 0);

    if (RARRAY(ary)->len == RHASH(hash)->tbl->num_entries) {
        return Qnil;
    }
    for (i=j=0; i<RARRAY(ary)->len; i++) {
        v = vv = rb_ary_elt(ary, i);
        if (st_delete(RHASH(hash)->tbl, (st_data_t*)&vv, 0)) {
            rb_ary_store(ary, j++, v);
        }
    }
    RARRAY(ary)->len = j;

    return ary;
}


[thomthom]

Jernej: how about larger iterations and higher number of random values?

[Nazz78]

Jernej: how about larger iterations and higher number of random values?

t=Time.now;a=Set.new;10000000.times{a.insert(rand(10000))};puts Time.now - t
Result: 37.911
t=Time.now;a=[];10000000.times{r=rand(10000);a<<r};a.uniq!; puts Time.now - t
Result: 8.282

Still a winner?

[thomthom]

It's refactoring time!

Nice find!

[RickW]

That's all great (using .uniq!) until you start dealing with Point3d objects
In that case, always use Set.