float(Union{Missing, Float32}) Failing in 1.0.0, From Using quantile In a Particular Way

[acrosby]

I have been trying to run quantile for an array of Array{Union{Missing, Float32}}, but that only contains valid Float32 values. I assume this should be possible, but I get the following error that seems to be related to float conversion:

From worker 3:	MethodError: no method matching AbstractFloat(::Type{Union{Missing, Float32}})
From worker 3:	Closest candidates are:
From worker 3:	  AbstractFloat(!Matched::Bool) at float.jl:250
From worker 3:	  AbstractFloat(!Matched::Int8) at float.jl:251
From worker 3:	  AbstractFloat(!Matched::Int16) at float.jl:252
From worker 3:	  ...
From worker 3:	float(::Type) at ./float.jl:269
From worker 3:	#quantile!#46(::Bool, ::Function, ::Array{Union{Missing, Float32},1}, ::Array{Float64,1}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.0/Statistics/src/Statistics.jl:819
From worker 3:	(::getfield(Statistics, Symbol("#kw##quantile!")))(::NamedTuple{(:sorted,),Tuple{Bool}}, ::typeof(Statistics.quantile!), ::Array{Union{Missing, Float32},1}, ::Array{Float64,1}) at ./none:0
From worker 3:	#quantile#52(::Bool, ::Function, ::Array{Union{Missing, Float32},1}, ::Array{Float64,1}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.0/Statistics/src/Statistics.jl:913
From worker 3:	quantile(::Array{Union{Missing, Float32},1}, ::Array{Float64,1}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.0/Statistics/src/Statistics.jl:913

Version:

julia> versioninfo()
Julia Version 1.0.0
Commit 5d4eaca0c9 (2018-08-08 20:58 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, ivybridge)

I know that the documentation suggests skipmissing which must convert to a standard Array{Float}, but I assume that the above should also be possible. Would adding skipmissing in my case add additional overhead, and is this any different from explicitly converting the with convert(Array{Float32}, ...) in my case?

Thanks!

[acrosby]

The minimal example to reproduce sheds some more light on the situation:

julia> using Statistics

julia> test = Array{Union{Missing, Float32},1}(undef, 2)
2-element Array{Union{Missing, Float32},1}:
 missing
 missing

julia> test[:] .= 2.
2-element view(::Array{Union{Missing, Float32},1}, :) with eltype Union{Missing, Float32}:
 2.0f0
 2.0f0

julia> quantile(test)
ERROR: MethodError: no method matching quantile(::Array{Union{Missing, Float32},1})
Closest candidates are:
  quantile(::AbstractArray{T,1} where T, ::Any; sorted) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.0/Statistics/src/Statistics.jl:913
  quantile(::Any, ::Any; sorted) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.0/Statistics/src/Statistics.jl:911
Stacktrace:
 [1] top-level scope at none:0

julia> float(test[1])
2.0f0

julia> quantile(test, 0.1)
2.0

julia> quantile(test, 0.5)
2.0

julia> quantile(test, [0.1, 0.5])
ERROR: MethodError: no method matching AbstractFloat(::Type{Union{Missing, Float32}})
Closest candidates are:
  AbstractFloat(::Bool) at float.jl:250
  AbstractFloat(::Int8) at float.jl:251
  AbstractFloat(::Int16) at float.jl:252
  ...
Stacktrace:
 [1] float(::Type) at ./float.jl:269

[nalimilan]

We should define float(::Type{Union{T,Missing}}) where {T} = float(T). That seems to be enough to fix it. Would you make a PR?

[acrosby]

Sure, I can take a crack at the process given that it may be such a simple fix. Cheers

[acrosby]

Actually, I'm not sure if this will have undesired downstream impacts, either now or in the future--Since float(x) is propagating the type union, and already works out of the context of quantile.

julia> test = Array{Union{Missing, Float32},1}(undef, 2)
2-element Array{Union{Missing, Float32},1}:
 missing
 missing

julia> float(test)
2-element Array{Union{Missing, Float32},1}:
 missing
 missing

julia> float(test[1])
missing

julia> test[:] .= 2.
2-element view(::Array{Union{Missing, Float32},1}, :) with eltype Union{Missing, Float32}:
 2.0f0
 2.0f0

julia> float(test)
2-element Array{Union{Missing, Float32},1}:
 2.0f0
 2.0f0

julia> float(test[1])
2.0f0

Will have to investigate further.

[nalimilan]

float(::Array{Union{T,Missing}}) where {T} works, but not float(::Type{Union{T,Missing}}) where {T}

[tpapp]

@nalimilan: I just came across this issue. Are you suggesting that fixing this could be as simple as adding that method to float?

[nalimilan]

Yes, probably.