Automatic inverse link

[palday]

• adds a way to create extensions for automatic inverse link determination
• defines such extensions for GLM and MixedModels
• this also allows using analytic deritvatives instead of AD for the difference method

[codecov]

Codecov Report

Merging #56 (e838625) into main (1cd621d) will decrease coverage by 0.58%.
The diff coverage is 95.65%.

@@             Coverage Diff             @@
##              main      #56      +/-   ##
===========================================
- Coverage   100.00%   99.42%   -0.58%     
===========================================
  Files            4        6       +2     
  Lines          154      174      +20     
===========================================
+ Hits           154      173      +19     
- Misses           0        1       +1     

Files Changed	Coverage Δ
src/Effects.jl	100.00% <ø> (ø)
src/regressionmodel.jl	97.77% <94.44%> (-2.23%)	⬇️
ext/EffectsGLMExt.jl	100.00% <100.00%> (ø)
ext/EffectsMixedModelsExt.jl	100.00% <100.00%> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[ararslan]

I think this can be simplified considerably by defining more generics directly in Effects that will then be extended when loading GLM or any package that depends on it.

[ararslan]

IIUC this will only be loaded if both MixedModels and GLM are loaded. I don't think that restriction is necessary since MixedModels depends on GLM.

[palday]

🤷 I wanted to be explicit

[ararslan]

Suggested change

	@test all(pred.prediction .≈ eff.vol)
	@test all(isapprox.(pred.lower, eff.lower; atol=0.001))
	@test all(isapprox.(pred.upper, eff.upper; atol=0.001))
	@test pred.prediction ≈ eff.vol
	@test pred.lower ≈ eff.lower atol=0.001
	@test pred.upper ≈ eff.upper atol=0.001

Any reason not to test the approximate equality of the arrays directly?

[palday]

because we want element-wise approximate equality, which is a stronger condition (the approximate equality of vectors is based on the norm of the difference vector)

[ararslan]

Suggested change

	@test all(eff_trans.vol .≈ eff.vol)
	@test all(isapprox.(eff_trans.lower, eff.lower; atol=0.001))
	@test all(isapprox.(eff_trans.upper, eff.upper; atol=0.001))
	@test eff_trans.vol ≈ eff.vol
	@test eff_trans.lower ≈ eff.lower atol=0.001
	@test eff_trans.upper ≈ eff.upper atol=0.001

Same question as above

[palday]

same answer as above