add hasintercept

[matthieugomez]

This is useful to compute the correct degrees of freedoms to be used in t-test

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (e42a88a) to head (8aa44b4).
⚠️ Report is 25 commits behind head on main.

Additional details and impacted files

@@ Coverage Diff @@## main #17 +/- ##
=========================================
Coverage 100.00% 100.00% =========================================
Files 2 3 +1 Lines 36 40 +4 =========================================
+ Hits 36 40 +4 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[nalimilan]

Thanks. Maybe this should be defined only for RegressionModel instead?

I also wonder whether we should even define a fallback definition X = modelmatrix(model); any(i -> all(==(1), view(X , :, i)), 1:size(X, 2)) like in GLM.jl.

[nalimilan]

Bump.

[matthieugomez]

Moved it to RegressionModel

[matthieugomez]

Btw, note that StatsModels started exporting hasintercept after I wrote this PR. Not sure if this makes it better or not to define it in StatsAPI

[nalimilan]

How about adding a fallback definition as I proposed above?

[nalimilan]

Btw, note that StatsModels started exporting hasintercept after I wrote this PR. Not sure if this makes it better or not to define it in StatsAPI

Ah, good point, that's JuliaStats/StatsModels.jl#281. I guess it makes sense to define it in StatsAPI too like the rest of the RegressionModel API, but we will have to import it in StatsModels after tagging a release. AFAICT StatsModels doesn't define any method for RegressionModel, only for terms/formulas.

[nalimilan]

Thinking about it, it could be better to only have an empty definition here, so that we can define a fallback in StatsModels like this:

function StatsModels.hasintercept(m::RegressionModel)
ifformula(m) !==nothingreturnhasintercept(formula(m))
else
X =modelmatrix(model)
returnany(i ->all(==(1), view(X , :, i)), 1:size(X, 2))
endend

EDIT: a third possibility would be to add a fallback formula(m::StatisticalModel) = nothing in StatsAPI so that we can have the method above in StatsAPI.

CC: @kleinschmidt@palday

[palday]

@nalimilan you need requires_one_based_indexing for yours or change it to use axes.

I like the idea of defining a formula stub here. Then downstream packages can do things like define RegressionModels that have a reasonable default formula definition without depending on StatsModels and still optionally supporting the whole Terms interface in extensions.

[nalimilan]

@kleinschmidt What do you think? I find it relatively unlikely that a package would want to provide an alternative formula implementation. Now that we've decided to put vif and gvif in StatsModels, maybe we should be consistent and put the fallback hasintercept definition there...