[swift-evolution] [Proposal] Random Unification

[Jens Persson]

I agree.

If there should be an API to produce a random Double without parameters
then IMHO it should simply be a uniformly distributed Double in the unit
range [0, 1). Very useful as a basic building block and can be constructed
very fast from a random bitpattern like eg:
extension Double {
    init(unitRange v: UInt64) {
        self = Double(v >> Double.exponentBitCount) * (.ulpOfOne/2.0)
    }
}



On Wed, Jan 10, 2018 at 11:48 PM, Saagar Jha <saagar at saagarjha.com> wrote:

> Which begs the question: why would you want to do something like this?
> Creating a “random” Double from its full range of values is an odd thing to
> do, and the fact that it is non-uniform and has irregularities like
> infinity and signed zeros makes it likely that any such usage is probably
> done in error (the one reason I can think of is if you’re trying to test
> your code with random Doubles, but I’d argue for keeping NaN in that case).
> Generally users are looking for uniform distributions or ones that follow
> some set pattern (e.g. normal, Bernoulli), so this doesn’t seem useful at
> all.
>
> Saagar Jha
>
> On Jan 10, 2018, at 14:39, Jens Persson <jens at bitcycle.com> wrote:
>
> On Wed, Jan 10, 2018 at 11:27 PM, Saagar Jha <saagar at saagarjha.com> wrote:
>
>> Not a floating point expert, but are you sure this works? I have a
>> feeling this would lead to a distribution that’s not uniform.
>>
>>
> Yes, it would not be uniform, which is exactly what I meant by the last
> part of: "Assuming you are ok with signed zero and infinities and
> "strange" bias as result of IEEE 754"
>
> Also, I think it's impossible to get a uniform distribution of all non-Nan
> Double values (since they include +- infinity).
>
> /Jens
>
>
>
>> Saagar Jha
>>
>> On Jan 10, 2018, at 14:07, Jens Persson via swift-evolution <
>> swift-evolution at swift.org> wrote:
>>
>> On Tue, Jan 9, 2018 at 10:07 PM, Jonathan Hull via swift-evolution <
>> swift-evolution at swift.org> wrote:
>>
>>>
>>> One additional question.  How do you ergonomically get a Double which
>>> doesn’t have a range, but also isn’t NaN?
>>>
>>>
>> Assuming you are ok with signed zero and infinities and "strange" bias as
>> result of IEEE 754:
>>
>> func randomNonNanDouble<R: RandomNumberGenerator>(using generator: R) ->
>> Double {
>>     while true {
>>         let rndUInt64 = generator.next()
>>         let rndDouble = Double(bitPattern: rndUInt64)
>>         if rndDouble != Double.nan { return rndDouble }
>>     }
>> }
>>
>> /Jens
>>
>> _______________________________________________
>> swift-evolution mailing list
>> swift-evolution at swift.org
>> https://lists.swift.org/mailman/listinfo/swift-evolution
>>
>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20180111/8677ced9/attachment.html>