Add dof_aov, dof_aovres, formula_aov

Author: yufongpeng

docs/src/Algorithm_AnovaFixedEffectModels.md

@@ -46,7 +46,7 @@ The mean residual deviance $\sigma^2$
 ```math
 \sigma^2 =\frac{D_n}{df_r}
 ```
-where $D_n$ is the residual sum of squares of $M_n$; $df_r$ is the degrees of freedom of the residuals, i.e. $df_r = s - n(\mathcal{C})$, where $s$ is number of samples.
+where $D_n$ is the residual sum of squares of $M_n$; $df_r$ is the degrees of freedom of the residuals, i.e. $df_r = nob - n(\mathcal{C})$, where $nob$ is number of observations.
 
 ## F-test
 F-value is a vector

docs/src/Algorithm_AnovaGLM.md

@@ -53,7 +53,7 @@ For ordinary linear regression,
 ```math
 \sigma^2 =\frac{D_n}{df_r}
 ```
-where $D_n$ is the residual sum of squares of $M_n$; $df_r$ is the degrees of freedom of the residuals, i.e. $df_r = s - n(\mathcal{C})$, where $s$ is number of samples.
+where $D_n$ is the residual sum of squares of $M_n$; $df_r$ is the degrees of freedom of the residuals, i.e. $df_r = nob - n(\mathcal{C})$, where $nob$ is number of observations.
 
 ## F-test
 F-value is a vector

docs/src/AnovaBase.md

@@ -44,6 +44,9 @@ AnovaBase.canonicalgoodnessoffit
 AnovaBase.ftest_nested
 AnovaBase.lrt_nested
 AnovaBase.dof_residual(aov::AnovaResult)
+AnovaBase.dof_aovres(::RegressionModel)
+AnovaBase.dof_aov(::RegressionModel)
+AnovaBase.formula_aov(::RegressionModel)
 AnovaBase.predictors(::RegressionModel)
 AnovaBase.anovatable(::AnovaResult{<: FullModel})
 ```

docs/src/Interface.md

@@ -23,7 +23,7 @@ It is recommended that `anova` additionally dispatched on wrapper types and the
 ## [predictors](./AnovaBase.md#AnovaBase.predictors-Tuple{RegressionModel})
 This function returns a tuple of terms which can be used in ANOVA (some terms may not be used because of ANOVA type or model itself).
 
-There is a default method for `RegressionModel`, i.e., `formula(model).rhs.terms`. If the formula for `SomeModel` has special structure like `MixedModel`, this function should be overloaded. 
+There is a default method for `RegressionModel`, i.e., `formula_aov(model).rhs.terms`. If the formula for `SomeModel` has special structure like `MixedModel`, this function or [`formula_aov`](./AnovaBase.md#AnovaBase.formula_aov-Tuple{RegressionModel}) should be overloaded. 
 
 ## AnovaModels
 ### [NestedModels](./AnovaBase.md#AnovaBase.NestedModels)
@@ -62,4 +62,7 @@ This function is not essential for ANOVA; it is just for convenience to create n
 `AnovaBase` provides a lot of functions to work on formula, terms and contrasts. See [Developer utility](./AnovaBase.md#Developer-utility)
 
 ## Other function
-* [`dof_residual`](./AnovaBase.md#StatsAPI.dof_residual-Tuple{AnovaResult}) applies `dof_residual` to all models by default. If `dof_residual(::SomeModel)` is not valid for ANOVA, customize `dof_residual(::AnovaResult{<: AnovaModel{SomeModel}})` alternatively.
+* [`dof_aovres`](./AnovaBase.md#AnovaBase.dof_aovres-Tuple{RegressionModel}) calls `dof_residual` by default. If `dof_residual(::SomeModel)` is not valid for ANOVA, customize `dof_aovres(::SomeModel)`. This function is utilized in the following function [`dof_residual`].
+* [`dof_residual`](./AnovaBase.md#StatsAPI.dof_residual-Tuple{AnovaResult}) applies [`dof_aovres`](./AnovaBase.md#AnovaBase.dof_aovres-Tuple{RegressionModel}) to all models by default. This function is utilized in the [`anovatable`](./AnovaBase.md#AnovaBase.anovatable-Tuple{AnovaResult{<:FullModel}}). 
+* [`dof_aov`](./AnovaBase.md#AnovaBase.dof_aov-Tuple{RegressionModel}) calls `dof` by default. If `dof(::SomeModel)` is not valid for ANOVA, customize `dof_aov(::SomeModel)`. Currently, no other function depends on this function.
+* [`formula_aov`](./AnovaBase.md#AnovaBase.formula_aov-Tuple{RegressionModel}) calls `formula` by default. If `formula(::SomeModel)` is not valid for ANOVA, customize `formula_aov(::SomeModel)`. Several functions including [`predictors`](./AnovaBase.md#AnovaBase.predictors-Tuple{RegressionModel}), and `show` function for ANOVA depend on this function. 

src/fit.jl

@@ -50,9 +50,9 @@ end
 
 # Calculate dof from assign
 """
-    dof_asgn(v::Vector{Int})
+    dof_asgn(assign::Vector{Int})
 
-Calculate degrees of freedom of each predictors. 'assign' can be obtained by `StatsModels.asgn(f::FormulaTerm)`. For a given `trm::RegressionModel`, it is as same as `trm.mm.assign`.
+Calculate degrees of freedom of each predictors. `assign` can be obtained by `StatsModels.asgn(f::FormulaTerm)`. For a given `trm::RegressionModel`, it is as same as `trm.mm.assign`.
 
 The index of the output matches values in the orinal `assign`. If any index value is not in `assign`, the default is 0.
 
@@ -80,8 +80,6 @@ function dof_asgn(v::Vector{Int})
     dofv
 end
 
-@deprecate dof(v::Vector{Int}) dof_asgn(v::Vector{Int})
-
 const FixDispDist = Union{Bernoulli, Binomial, Poisson}
 """
     canonicalgoodnessoffit(::FixDispDist) = LRT

src/interface.jl

@@ -46,20 +46,47 @@ end
 
 Degrees of freedom of residuals.
 
-By default, it applies `dof_residual` to models in `aov.anovamodel`.
+By default, it applies `dof_aovres` to models in `aov.anovamodel`.
 """
-dof_residual(aov::AnovaResult{M, T, N}) where {M, T, N} = ntuple(x -> dof_residual(aov.anovamodel.model), N)
-dof_residual(aov::AnovaResult{<: MultiAovModels}) = map(dof_residual, aov.anovamodel.model)
+dof_residual(aov::AnovaResult{M, T, N}) where {M, T, N} = ntuple(x -> dof_aovres(aov.anovamodel.model), N)
+dof_residual(aov::AnovaResult{<: MultiAovModels}) = map(dof_aovres, aov.anovamodel.model)
+
+"""
+    dof_aovres(m::RegressionModel)
+
+Degrees of freedom of residuals for ANOVA.
+
+By default, it calls `dof_residual`.
+"""
+dof_aovres(m::RegressionModel) = dof_residual(m)
+
+"""
+    dof_aov(m::RegressionModel)
+
+Degrees of freedom of model for ANOVA.
+
+By default, it calls `dof`.
+"""
+dof_aov(m::RegressionModel) = dof(m)
+
+"""
+    formula_aov(m::RegressionModel)
+
+Formula of model for ANOVA.
+
+By default, it calls `formula`.
+"""
+formula_aov(m::RegressionModel) = formula(m)
 
 """
     predictors(model::RegressionModel)
     predictors(anovamodel::FullModel)
 
 Return a tuple of `Terms` which are predictors of the model or anovamodel. 
 
-By default, it returns `formula(model).rhs.terms`; if the formula has special structures, this function should be overloaded.
+By default, it returns `formula_aov(model).rhs.terms`; if the formula has special structures, this function should be overloaded.
 """
-predictors(model::RegressionModel) = formula(model).rhs.terms
+predictors(model::RegressionModel) = formula_aov(model).rhs.terms
 predictors(anovamodel::FullModel) = getindex.(Ref(predictors(anovamodel.model)), anovamodel.pred_id)
 
 """

src/io.jl

@@ -10,8 +10,6 @@ testname(::Type{FTest}) = "F test"
 testname(::Type{LRT}) = "Likelihood-ratio test"
 #testname(M::AnovaStatsRao) = "Rao score test"
 #testname(M::AnovaStatsCp) = "Mallow's Cp"
-@deprecate tname testname
-
 """
     prednames(aov::AnovaResult)
     prednames(anovamodel::FullModel) 
@@ -30,17 +28,14 @@ function prednames(anovamodel::MultiAovModels)
 end
 prednames(model::RegressionModel) = vectorize(prednames(predictors(model)))
 
-@deprecate coefnames(aov::AnovaResult) prednames(aov::AnovaResult)
-@deprecate coefnames(x, ::Val{:anova}) prednames(x)
-
 function show(io::IO, anovamodel::FullModel)
     println(io, "FullModel for type $(anovamodel.type) test")
     println(io)
     println(io, "Predictors:")
     println(io, join(prednames(anovamodel), ", "))
     println(io)
     println(io, "Formula:")
-    println(io, formula(anovamodel.model))
+    println(io, formula_aov(anovamodel.model))
     println(io)
     println(io, "Coefficients:")
     show(io, coeftable(anovamodel.model))
@@ -51,7 +46,7 @@ function show(io::IO, anovamodel::NestedModels{M, N}) where {M, N}
     println(io)
     println(io, "Formulas:")
     for(id, m) in enumerate(anovamodel.model)
-        println(io, "Model $id: ", formula(m))
+        println(io, "Model $id: ", formula_aov(m))
     end
     println(io)
     println(io, "Coefficients:")
@@ -66,7 +61,7 @@ function show(io::IO, anovamodel::MixedAovModels{M, N}) where {M, N}
     println(io)
     println(io, "Formulas:")
     for(id, m) in enumerate(anovamodel.model)
-        println(io, "Model $id: ", formula(m))
+        println(io, "Model $id: ", formula_aov(m))
     end
     println(io)
     println(io, "Coefficients:")
@@ -83,7 +78,7 @@ function show(io::IO, aov::AnovaResult{<: FullModel, T}) where {T <: GoodnessOfF
     println(io)
     println(io, "Type $(anova_type(aov)) test / $(testname(T))")
     println(io)
-    println(io, formula(aov.anovamodel.model))
+    println(io, formula_aov(aov.anovamodel.model))
     println(io)
     println(io, "Table:")
     show(io, at)
@@ -96,7 +91,7 @@ function show(io::IO, aov::AnovaResult{<: MultiAovModels, T}) where {T <: Goodne
     println(io, "Type $(anova_type(aov)) test / $(testname(T))")
     println(io)
     for(id, m) in enumerate(aov.anovamodel.model)
-        println(io, "Model $id: ", formula(m))
+        println(io, "Model $id: ", formula_aov(m))
     end
     println(io)
     println(io, "Table:")

src/term.jl

@@ -1,6 +1,4 @@
 # Function related to terms
-@deprecate has_intercept hasintercept
-
 """
     any_not_aliased_with_1(terms)
 
@@ -98,8 +96,7 @@ getterms(term::FunctionTerm) = [term.exorig]
 
 # Determine selected terms for type 2 ss
 """
-    isinteract(m::MatrixTerm, id1::Int, id2::Int)
-    isinteract(f::TupleTerm, id1::Int, id2::Int)
+    isinteract(f::Union{MatrixTerm, TupleTerm}, id1::Int, id2::Int)
 
 Determine if `f[id2]` is an interaction term of `f[id1]` and other terms.
 
@@ -133,14 +130,10 @@ isinteract(f::MatrixTerm, id1::Int, id2::Int) = isinteract(f.terms, id1, id2)
 isinteract(f::TupleTerm, id1::Int, id2::Int) = issubset(getterms(f[id1]), getterms(f[id2]))
 
 const doc_select_interaction = """
-    select_super_interaction(m::MatrixTerm, id::Int)
-    select_super_interaction(f::TupleTerm, id::Int)
-    select_sub_interaction(m::MatrixTerm, id::Int)
-    select_sub_interaction(f::TupleTerm, id::Int)
-    select_not_super_interaction(m::MatrixTerm, id::Int)
-    select_not_super_interaction(f::TupleTerm, id::Int)
-    select_not_sub_interaction(m::MatrixTerm, id::Int)
-    select_not_sub_interaction(f::TupleTerm, id::Int)
+    select_super_interaction(f::Union{MatrixTerm, TupleTerm}, id::Int)
+    select_sub_interaction(f::Union{MatrixTerm, TupleTerm}, id::Int)
+    select_not_super_interaction(f::Union{MatrixTerm, TupleTerm}, id::Int)
+    select_not_sub_interaction(f::Union{MatrixTerm, TupleTerm}, id::Int)
 
 Return a set of index of `f`, which
 
@@ -220,8 +213,6 @@ function select_not_sub_interaction(f::TupleTerm, id::Int)
     s
 end
 
-@deprecate selectcoef(f::MatrixTerm, id::Int) select_super_interaction(f, id)
-
 # Create sub-formula
 """
     subformula(f::FormulaTerm, id; kwargs...)
@@ -331,8 +322,6 @@ function clear_schema(t::MatrixTerm)
     length(ts) ≡ 1 ? ts[1] : ts
 end
 
-@deprecate clearschema clear_schema
-
 # reschema only happen when using TupleTerm rather than MatrixTerm
 reschema_formula(lhs::AbstractTerm, ts::TupleTerm, reschema::Bool) = 
     reschema ? FormulaTerm(clear_schema(lhs), clear_schema.(ts)) : FormulaTerm(lhs, collect_matrix_terms(ts))

test/runtests.jl

@@ -1,5 +1,5 @@
 using AnovaBase
-import AnovaBase: dof_residual, nobs, anovatable, prednames, predictors, coeftable, dof_asgn, canonicalgoodnessoffit
+import AnovaBase: dof_residual, dof_aovres, formula_aov, dof_aov, dof, nobs, anovatable, prednames, predictors, coeftable, dof_asgn, canonicalgoodnessoffit
 using Distributions: Gamma, Binomial
 import Base: show
 using Test
@@ -28,11 +28,12 @@ macro test_error(x)
     end
 end
 
+dof(x::Int) = x
 dof_residual(x::Int) = x
 nobs(x) = ntuple(one, length(x))
 nobs(x::Int) = one(x)
 predictors(::Int) = tuple(Term.(Symbol.(["x$i" for i in 1:7]))...)
-predictors(model::StatsModels.TableRegressionModel{Int64, Matrix{Float64}}) = formula(model).rhs
+predictors(model::StatsModels.TableRegressionModel{Int64, Matrix{Float64}}) = formula_aov(model).rhs
 coeftable(model::StatsModels.TableRegressionModel{Int64, Matrix{Float64}}) = []
 anovatable(::AnovaResult{<: FullModel{StatsModels.TableRegressionModel{Int64, Matrix{Float64}}}, LikelihoodRatioTest, 7}; rownames = string.(1:7)) = 
     AnovaBase.AnovaTable([
@@ -168,9 +169,13 @@ anovatable(::AnovaResult{<: FullModel{StatsModels.TableRegressionModel{Int64, Ma
                                 ntuple(one ∘ float, 3),
                                 ntuple(zero ∘ float, 3),
                                 NamedTuple())
-    @testset "attr.jl" begin
+    @testset "attr.jl and interface.jl" begin
         @test nobs(ft) == 1
         @test nobs(lrt) == 1
+        @test dof_aov(ft.anovamodel.model[1]) == 1
+        @test dof_aov(lrt.anovamodel.model) == 1
+        @test formula_aov(ft.anovamodel.model[1]) == formula(ft.anovamodel.model[1])
+        @test formula_aov(lrt.anovamodel.model) == formula(lrt.anovamodel.model)
         @test dof_residual(lrt) == (1, 1, 1, 1, 1, 1, 1)
         @test dof_residual(lrt2) == (1, 1, 1)
         @test deviance(ft) == ft.deviance