[ruby-core:34219] Re: [Ruby 1.9-Feature#4247][Open] New features for Array#sample, Array#choice

Hi,

On 7 January 2011 12:08, Yusuke ENDOH <[e-mail]> wrote:
> Hi,
>
> 2011/1/7 Yoji Ojima <[e-mail]>:
>> We are planning to add the following features of the random sampling to Array.
>>
>> 1. Weighted random sampling.
>> 2. Sampling with replacement.
>> 3. Iteration.
>>
>> It is discussed in ruby-dev (Feature #3647 and #4147).
>
>
> I'm adding complementary information.
>
> There are two reasons why the name "choice" is selected:
>
>  For backward compatibility. nce upon a time, trunk (before 1.9.0)
> ad provided Array#choice. t was backported to 1.8.7. hen, in
> runk, the name was changed to Array#sample (see the thread from
> ruby-core:18036]). ut 1.8.7 still provides Array#choice because
> t cannot remove Array#choice for compatibility reason.
> ote that 1.8.7's Array#choice does not receive any argument, so
> here is no compatibility problem.
>
>  Mathematica provides RandomSample and RandomChoice for SRSWOR and
> RSWR, respectively
> ttp://reference.wolfram.com/mathematica/ref/RandomSample.html
> ttp://reference.wolfram.com/mathematica/ref/RandomChoice.html
>
>
> There are some algorithms [1] [2] for fast multiple sampling.
>
> [1] Pavlos S. Efraimidis, Paul G. Spirakis
> eighted random sampling with a reservoir
> nformation Processing Letters
> olume 97, Issue 5 (16 March 2006)
> Ruby implementation is in [ruby-dev:42844])
>
> [2] A. J. Walker
> n Efficient Method for Generating Discrete Random Variables with
> General Distributions
> CM Transactions on Mathematical Software, 3 (1977), 253-256.
>
>
> Matz roughly approved this suggestion. ut he said that the method name
> of "each_sample" and "each_choice" are a bit awkward, and that he want
> to hear opinions of ruby-core folks.
> Of course, we appreciate any comments about the feature itself rather
> than the name.
>
> --
> Yusuke Endoh <[e-mail]>
>
>

Thanks for the name clarification.

I think 'choice' and 'each_choice' are weird.

Sample is "a subset of a population", and so it seems logical to have
multiple elements in return.
But choice seems like 'singular', only meant for one element. Is it
correct to say "Array#choice returns a choice of some random elements"
? It does not seems right to me.

To this idea, #choice should always return one element, and #sample
could be the enumerator (the form which returns an Array would then be
Array#sample.take(size)).
However, I guess that would break too much compatibility with current versions.
And I really like 1.9.2 name of #sample for 'a single random element'
(even if it might be incorrect to statistics).

Did you consider having another option to {each_,}sample to allow replacement ?
Such as:
ary.sample 3, replace: true

I see the title of the feature is "Array#sample(n, replace=false)", so
I guess the idea was there. How was it decided to instead go for
another set of methods ?
(Just because of Mathematica's choice ? Compatibility seems against
users' interest if it is to use a wrong method name)

Or to follow combinations/permutations:
ary.repeated_sample 3

In http://en.wikipedia.org/wiki/Simple_random_sample, the terms used
are "simple random sampling with/without replacement", that comfort me
to think choice is not the right .. choice (at least for multiple
elements).

I am not a native English speaker, so I might be not accurate about
this. If this is the case, please ignore what I said.


About the feature, I wonder if an Hash would be a good idea for the
weight option.
It can be very similar to a Proc with #default_proc, except much
faster for already stored values.


Off-topic:
I think it would be nice if some ruby-dev/ruby-core discussions could be merged.
Even if I cannot currently understand Japanese, I can at least read
their "coded" ideas.
I would see that as using the same topic, with both languages.