roadmap to 1.0 release

[kleinschmidt]

Current plan is that we'll tag 0.7 ASAP (today probably) and then we commit to tagging 1.0 in one month (so Feb 24). #183 is sufficiently breaking taht we want to make sure that dependencies have time to upgrade and shake out any sneaky bugs that may require further breakage before we commit to 1.0.

For 0.7:

• FunctionTerm is dead, long live FunctionTerm #183
• Export hasintercept and omitsintercept? #276

For 1.0: (with possible 0.8 if we want to give dependencies a release to target for upgrades)

• remove table regression model
• maybe WIP use Tables.Columns instead of columntable #247
• breaking changes to contrasts #175 Allow arbitrary storage for ContrastsMatrix and other contrasts breakage #273

[palday]

Packages we need to check (feel free to edit/add to)

• MixedModels.jl integrate changes from StatsModels 0.7 (upcoming release) MixedModels.jl#664
• Effects.jl
• GLM.jl add compat for statsmodels 0.7 GLM.jl#512
• RegressionFormulae.jl
• FixedEffectsModels.jl

[PharmCat]

Hi! Can you kindly clarify - can #154 be included in 1.0 release plan? I suppose apply_context stage can close some other issues as #268 ...

[PharmCat]

Here one PR I make for GLM.jl : JuliaStats/GLM.jl#508 for Testing of Between-Subjects Effects, there one problem: it relies very heavily on the internal structures of the types from StatsModels. But I didn't found in StatsModels appropriate functions with methods to access these quantities in a stable way for this code:

"""
    lcontrast(obj, i::Int)

L-contrast matrix for `i` fixed effect.
"""
function lcontrast(obj, i::Int)
    n = length(obj.mf.f.rhs.terms)
    if i > n || n < 1 error("Factor number out of range 1-$(n)") end
    p    = size(obj.mm.m, 2) # number of coefs
    inds = findall(x -> x==i, obj.mm.assign)
    if typeof(obj.mf.f.rhs.terms[i]) <: CategoricalTerm
        mxc   = zeros(size(obj.mf.f.rhs.terms[i].contrasts.matrix, 1), p)
        mxcv  = view(mxc, :, inds)
        mxcv .= obj.mf.f.rhs.terms[i].contrasts.matrix
        mx    = zeros(size(obj.mf.f.rhs.terms[i].contrasts.matrix, 1) - 1, p)
        for i = 2:size(obj.mf.f.rhs.terms[i].contrasts.matrix, 1) # correct for zero-intercept model
            mx[i-1, :] .= mxc[i, :] - mxc[1, :]
        end
    else
        mx = zeros(length(inds), p) # unknown correctness for zero-intercept model
        for j = 1:length(inds)
            mx[j, inds[j]] = 1
        end
    end
    mx
end

tname(t::AbstractTerm) = "$(t.sym)"
tname(t::InteractionTerm) = join(tname.(t.terms), " & ")
tname(t::InterceptTerm) = "(Intercept)"

[kleinschmidt]

Hi! Can you kindly clarify - can #154 be included in 1.0 release plan? I suppose apply_context stage can close some other issues as #268 ...

nope, not on the roadmap. the goal here is to get 1.0 released with essentially what is one master right now and with whatever other simple but breaking changes we've had planned/ready for a while.

#154 is much more speculative and I'm not really convinced it's a good idea still (I think the problem is real, but not sure that that's the solution 😁)

[matthieugomez]

Great to see this update! FixedEffectModels.jl has a PR for v0.7

[PharmCat]

Current plan is that we'll tag 0.7 ASAP (today probably)...

For 0.7:

* [x]  [FunctionTerm is dead, long live FunctionTerm #183](https://github.com/JuliaStats/StatsModels.jl/pull/183)

* [ ]  [Export hasintercept and omitsintercept? #276](https://github.com/JuliaStats/StatsModels.jl/issues/276)

Hi! Is any problem to tag 0.7 now?

[kleinschmidt]

0.7 registration PR opened 🎉 JuliaRegistries/General#79527

[kleinschmidt]

...and merged.