Regression should accept vectors

[tbreloff]

I was surprised that this wasn't allowed:

julia> using OnlineStats

julia> LinReg(randn(100),randn(100))
ERROR: MethodError: `convert` has no method matching convert(::Type{OnlineStats.LinReg{W<:OnlineStats.Weighting}}, ::Array{Float64,1}, ::Array{Float64,1})
This may have arisen from a call to the constructor OnlineStats.LinReg{W<:OnlineStats.Weighting}(...),
since type constructors fall back to convert methods.
Closest candidates are:
  OnlineStats.LinReg(::AbstractArray{Float64,2}, ::AbstractArray{Float64,1})
  OnlineStats.LinReg(::AbstractArray{Float64,2}, ::AbstractArray{Float64,1}, ::OnlineStats.Weighting)
  call{T}(::Type{T}, ::Any)
  ...
 in call at essentials.jl:57

julia> LinReg(randn(100,1),randn(100))
OnlineStat: OnlineStats.LinReg{OnlineStats.EqualWeighting}
 *        β:  [0.06140403868184703]
 *     nobs:  100

We should make sure both vectors and matrices are handled as expected.

[joshday]

I'm not completely sure this is unexpected. A quick survey shows the following require x to be a Matrix:

GLM.lm(x,y)
Regression.linearreg(x,y)
GLMNet.glmnet(x,y) 
MultivariateStats.llsq(x,y)

This also only happens if you're fitting a line without an intercept, since LinReg doesn't add one for you, although maybe it should.

[joshday]

I'd rather force users into doing x = view(randn(100), :, :) or similar. I think the issue is rare enough that I'd rather not add extra constructor/fit! methods for LinReg, QuantReg, and StatLearn.