lm(X,y) supports y as integer, but not X

[setzler]

Currently, lm(X,y) has asymmetric type support - y can be integer, X cannot. Please permit X as integer type. Example of the issue:

julia> y
3-element Array{Int64,1}:
2
4
6

julia> X
3x1 Array{Int64,2}:
1
2
3

julia> lm(X,y)
ERROR: DensePredQR{T<:Union(Float64,Float32)} has no method matching DensePredQR{T<:Union(Float64,Float32)}(::Array{Int64,2})

julia> X
3x1 Array{Float64,2}:
1.0
2.0
3.0

julia> lm(X,y)
LinearModel{DensePredQR{Float64}}:

Coefficients: Estimate Std.Error t value Pr(>|t|)
x1 2.0 3.84593e-16 5.20031e15 <1e-31

[pkofod]

This issue is now different.

 lm(rand(Int64, 4,2), rand(Int64, 4))
 lm(rand(4,2), rand(Int64, 4))
 lm(rand(Int64, 4,2), rand(4))

Neither work, but this works

 lm(rand(4,2), rand(4))

So either this issue is solved by requiring all floats, or it has expanded and now y can't be Integer either. This is in contrast to the fact that all these work

 glm(rand(Int64, 4,2), rand(Int64, 4), Normal(), IdentityLink())
 glm(rand(4,2), rand(Int64, 4), Normal(), IdentityLink())
 glm(rand(Int64, 4,2), rand(4), Normal(), IdentityLink())
 glm(rand(4,2), rand(4), Normal(), IdentityLink())

[nalimilan]

This works because fit for AbstractGLM has a generic method which calls float on the two vectors. Should be easy to add for linear models too.

[pkofod]

I see it. I'll make a PR harmonizing glm and lm fit methods.