Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add keyword arg to modelmatrix; define momentmatrix #16

Open
wants to merge 19 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from 12 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 17 additions & 9 deletions src/regressionmodel.jl
Original file line number Diff line number Diff line change
Expand Up @@ -34,19 +34,25 @@ Return the mean of the response.
function meanresponse end

"""
modelmatrix(model::RegressionModel)
modelmatrix(model::RegressionModel; weighted::Bool=false)

Return the model matrix (a.k.a. the design matrix).
Return the model matrix (design matrix) or, if `weighted=true` the weighted
model matrix, i.e. `X * sqrt.(W)`, where `X` is the model matrix and
`W` is the diagonal matrix whose elements are the [model weights](@ref weights(::StatisticalModel)).
"""
function modelmatrix end
function modelmatrix(model::RegressionModel; weighted::Bool=false) end
gragusa marked this conversation as resolved.
Show resolved Hide resolved

"""
crossmodelmatrix(model::RegressionModel)
crossmodelmatrix(model::RegressionModel; weighted::Bool=false)

Return `X'X` where `X` is the model matrix of `model`.
Return `X'X` where `X` is the model/design matrix of `model` or, if `weighted=true`, `X'WX`,
where `W` is the diagonal matrix whose elements are the [model weights](@ref weights(::StatisticalModel)).
This function will return a pre-computed matrix stored in `model` if possible.
"""
crossmodelmatrix(model::RegressionModel) = (x = modelmatrix(model); Symmetric(x' * x))
function crossmodelmatrix(model::RegressionModel; weighted::Bool=false)
x = weighted ? modelmatrix(model; weighted=weighted) : modelmatrix(model)
return Symmetric(x' * x)
end

"""
leverage(model::RegressionModel)
Expand All @@ -65,11 +71,13 @@ of each data point.
function cooksdistance end

"""
residuals(model::RegressionModel)
residuals(model::RegressionModel; weighted::Bool=false)

Return the residuals of the model or, if `weighted=true`, the residuals multiplied by
the square root of the [model weights](@ref weights(::StatisticalModel)).
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Where does the square root come from exactly? Doesn't that assume a particular definition of residuals (i.e. using L2-norm rather than e.g. L1-norm)?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Well, this is tricky. Also modelmatrix multiplies the entries of X by the square-root of the weights. Why?

Think about the linear model. With weights, the crossmodel matrix is $X'WX$. Then, to obtain it we can now do modelmatrix(lm1; weighted = true)'modelmatrix(lm1; weighted = true).

Notice that this is consistent with R; see, e.g., the function weighted.residuals which is in stats.

With weights, any weights is
$$\sqrt{w_i}y_i = \sqrt{w_i}x_i \beta + \sqrt{w_i}u_i.$$ So the understanding is that weighting single constituents of the model (y,x,u) amount to weight by $\sqrt{w_i}$.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, another tricky point. My understanding is that for residuals, the square root comes from the fact that deviance residuals themselves are defined as the square root of quantities which are partitions of the deviance. Right?

Note the R docstring for weighted.residuals says "Weighted residuals are based on the deviance residuals", which are only one kind of residual. Actually in R residuals also returns weighted residuals, except for response residuals, which are always unweighted. Maybe to be completely accurate we could say "for deviance and Pearson residuals...", so that packages are free to use different definitions (or throw an error) if needed?

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think what @nalimilan says is that the assumption here (and in your change of modelmatrix) is that for all kinds of weights the weighted model matrix is X * sqrt.(W). Is it always true for FrequencyWeights, AnalyticWeights and ProbabilityWeights? x-ref: JuliaStats/GLM.jl#487

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@bkamins weighted residuals, weighted model matrix do not exist in statistics. They are only useful from a coding point of view - they make it easier to write neater code.

I have always defined these quantities as multiplied by $\sqrt{w_i}$ as it is much more convenient. Some thing for R — which returns silently squared-root weighted residuals. Also other packages, notable FixedEffectModels.jl does that.

@nalimilan make sense what you propose - I will add more context to the doc

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Even if these don't exist in statistics, the question can be phrased as "are there situations where the returned value is useful, even when you don't know the kind of weights used". I think the answer is yes, but it's tricky, so... R base only supports analytic weights so it's not a great reference.


Return the residuals of the model.
"""
function residuals end
function residuals(model::RegressionModel; weighted::Bool=false) end

"""
predict(model::RegressionModel, [newX])
Expand Down
21 changes: 21 additions & 0 deletions src/statisticalmodel.jl
Original file line number Diff line number Diff line change
Expand Up @@ -314,3 +314,24 @@ function adjr2(model::StatisticalModel, variant::Symbol)
end

const adjr² = adjr2

"""
momentmatrix(model::StatisticalModel)
nalimilan marked this conversation as resolved.
Show resolved Hide resolved

Return the matrix containing the empirical moments defining the estimated parameters.

For likelihood-based models, `momentmatrix` returns the log-score, i.e. the gradient
of the log-likelihood evaluated at each observation. For semiparametric models, each row
of the ``(n \\times k)`` matrix returned by `momentmatrix` is ``m(Z_i, b)``, where `m`
is a vector-valued function evaluated at the estimated parameter `b` and ``Z_i`` is the
vector of data for entity `i`. The vector-valued function satisfies ``\\sum_{i=1}^n m(Z_i, b) = 0``.

For linear and generalized linear models, the parameters of interest are the coefficients
of the linear predictor. The moment matrix of a linear model is given by `u.*X`,
where `u` is the vector of residuals and `X` is the model matrix. The moment matrix of
a generalized linear model with link function `g`, the moment matrix is `X'e`, where `e`
gragusa marked this conversation as resolved.
Show resolved Hide resolved
is given by ``Y-g^{-1}(X'b)`` where `X` is the model matrix, `Y` is the model response, and
`b` is the vector of estimated coefficients.
"""
function momentmatrix end

18 changes: 18 additions & 0 deletions test/regressionmodel.jl
Original file line number Diff line number Diff line change
Expand Up @@ -6,13 +6,31 @@ using StatsAPI: RegressionModel, crossmodelmatrix
struct MyRegressionModel <: RegressionModel
end

struct MyWeightedRegressionModel <: RegressionModel
wts::AbstractVector
end

StatsAPI.modelmatrix(::MyRegressionModel) = [1 2; 3 4]

function StatsAPI.modelmatrix(r::MyWeightedRegressionModel; weighted::Bool=false)
X = [1 2; 3 4]
weighted ? sqrt.(r.wts).*X : X
end

w = [0.3, 0.2]

@testset "TestRegressionModel" begin
m = MyRegressionModel()
r = MyWeightedRegressionModel(w)

@test crossmodelmatrix(m) == [10 14; 14 20]
@test crossmodelmatrix(m; weighted=false) == [10 14; 14 20]
@test crossmodelmatrix(m) isa Symmetric

@test crossmodelmatrix(r) == [10 14; 14 20]
@test crossmodelmatrix(r; weighted=false) == [10 14; 14 20]
@test crossmodelmatrix(r; weighted=true) ≈ [2.1 3.0; 3.0 4.4]
@test crossmodelmatrix(r; weighted=true) isa Symmetric
end

end # module TestRegressionModel
2 changes: 1 addition & 1 deletion test/statisticalmodel.jl
Original file line number Diff line number Diff line change
Expand Up @@ -36,4 +36,4 @@ StatsAPI.nobs(::MyStatisticalModel) = 100
@test adjr2 === adjr²
end

end # module TestStatisticalModel
end # module TestStatisticalModel