Why doesn't rand work with AbstractFloat?

Question

In Julia 0.4.0, when I try

rand(AbstractFloat, 1)

The following error is obtained:

ERROR: MethodError: `rand` has no method matching rand(::MersenneTwister,
::Type{AbstractFloat})

Is there a reason behind the fact that I must explicitly say Float32 or Float64 for rand to work? Or is it just that, as the language is relatively new, a relevant method has yet to be defined in the Base?

Julia don't know which type you prefer among BigFloat, Float16, Float32 and Float64. — Gnimuc
– Gnimuc, Commented Oct 29, 2015 at 12:03
But for example one(AbstractFloat) gives either a Float32 or Float64 object depending on the system. Shouldn't rand have the same behaviour? — Taiki
– Taiki, Commented Oct 29, 2015 at 12:15
Good case, in my opinion, it's better for Julia to behave the same in alike situations, like the above. — Reza Afzalan
– Reza Afzalan, Commented Oct 29, 2015 at 13:07
Since rand(Float64) does not actually cover the whole range of Float64 values, but the [0,1) interval unlike the Int family, it seems reasonable to define this as an Abstract behavior and extend to AbstractFloat. Good point @Taiki — Dan Getz
– Dan Getz, Commented Oct 29, 2015 at 14:45

Gnimuc · Accepted Answer · 2015-10-29 15:31:32Z

4

one is different from rand. when using one(AbstractFloat), all the outputs are the "same":

julia> one(Float64)
1.0

julia> one(Float32)
1.0f0

julia> 1.0 == 1.0f0
true

this is not true when using rand:

julia> rand(srand(1), Float64)
0.23603334566204692

julia> rand(srand(1), Float32)
0.5479944f0

julia> rand(srand(1), Float32) == rand(srand(1), Float64)
false

this means if rand would behave like one, one might get two different results with the same seed on two different machines(e.g. one is x86, another is x64). take a look at the code in random.jl:

@inline rand{T<:Union{Bool, Int8, UInt8, Int16, UInt16, Int32, UInt32}}(r::MersenneTwister, ::Type{T}) = rand_ui52_raw(r) % T

both rand(Signed)&rand(Unsigned) are illegal too.

edited Oct 29, 2015 at 15:31

answered Oct 29, 2015 at 15:05

Gnimuc

8,6362 gold badges39 silver badges52 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Taiki Over a year ago

Very reasonable :) Thanks.

Collectives™ on Stack Overflow

Why doesn't rand work with AbstractFloat?

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related