DataFrame not defined error when including interval keyword argument

[mthelm85]

I'm getting a DataFrame not defined error when attempting to use the interval keyword argument in the predict function.

Code to reproduce:

using DataFrames, GLM

data = DataFrame(X=[1,2,3], Y=[2,4,7])
ols = lm(@formula(Y ~ X), data)
new_data = DataFrame(X=[5])

# Without the interval kwarg

julia> predict(ols, new_data)
1-element Array{Union{Missing, Float64},1}:
 11.833333333333343

# With the interval kwarg

julia> predict(ols, new_data, interval=:prediction)
ERROR: UndefVarError: DataFrame not defined
Stacktrace:
 [1] _return_predictions(::NamedTuple{(:prediction, :lower, :upper),Tuple{Array{Float64,1},Array{Float64,2},Array{Float64,2}}}, ::BitArray{1}, ::Int64) at C:\Users\mthel\.julia\packages\StatsModels\Kz7By\src\statsmodel.jl:153
 [2] #predict#74(::Base.Iterators.Pairs{Symbol,Symbol,Tuple{Symbol},NamedTuple{(:interval,),Tuple{Symbol}}}, ::typeof(predict), ::StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Array{Float64,1}},GLM.DensePredChol{Float64,LinearAlgebra.Cholesky{Float64,Array{Float64,2}}}},Array{Float64,2}}, ::DataFrame) at C:\Users\mthel\.julia\packages\StatsModels\Kz7By\src\statsmodel.jl:166
 [3] (::getfield(StatsBase, Symbol("#kw##predict")))(::NamedTuple{(:interval,),Tuple{Symbol}}, ::typeof(predict), ::StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Array{Float64,1}},GLM.DensePredChol{Float64,LinearAlgebra.Cholesky{Float64,Array{Float64,2}}}},Array{Float64,2}}, ::DataFrame) at .\none:0
 [4] top-level scope at REPL[11]:1

[nalimilan]

Good catch. @kleinschmidt you thought StatsModels didn't depend on DataFrames anymore, but you cheated!

It shouldn't be too hard to fix by using Tables.materializer on the input data to return the same type of table object, without depending on DataFrames. We would return a NamedTuple when a matrix is passed. I can fix this when getting rid of TableRegressionModel (#32), but we may want to apply a simpler fix before that.

Tests don't catch it probably because they load DataFrames. Designing a test that would catch the error sounds hard.

[nalimilan]

See #160 for the fix, and JuliaStats/GLM.jl#335 for new tests.