wip: recipes for regression models. should close #290 #293
Conversation
mkborregaard
force-pushed
the
mkb/statsmodels
branch
from
d980c7f to
e2bab1b
Compare
|
Nice. |
| @@ -0,0 +1,8 @@ | |||
| # Note both predict and RegressionModel are defined in StatsBase | |||
| @recipe function f(mod::RegressionModel) | |||
| newx = [ones(200) range(extrema(mod.model.pp.X[:,2])..., length = 200)] | |||
There was a problem hiding this comment.
Can we be sure that the second column is what we want to plot?
There was a problem hiding this comment.
Maybe this is a good hack for now?
@recipe function f(mod::RegressionModel, name)
newx = [ones(200) range(extrema(mod.model.pp.X[:,findfirst(==(name), coefnames(mod))])..., length = 200)]
....There was a problem hiding this comment.
So the interface being name can be a symbol, but also a Tuple or Vector of symbols, length, and then the plot being bivariate or 3d depending on that, and the regression keeping all the other input variables at their mean?
There was a problem hiding this comment.
I just meant that the second column isn't always going to be what we want to plot.
For example...
mod = lm(@formula(PetalLength ~ 0 + PetalWidth), iris)would get rid of the intercept so the 2nd column wouldn't even exist. I was just trying to find something that would ensure the right coefficient is grabbed from the model matrix.
There was a problem hiding this comment.
Yes I know and I agree, I was just trying to see if we could make it even more general
There was a problem hiding this comment.
That would actually be really convenient to be able to add multiple coefficients and see how they are interacting visually! I think this would also make for a good long term user interface when the stats ecosystem will adopt a better way of extracting information from models using coefficient names.
There was a problem hiding this comment.
mod.model.pp.X is specific to GLM, you should rather use modelmatrix(mod) instead.
Right now only works for the bivariate case
mkborregaard
force-pushed
the
mkb/statsmodels
branch
from
d232496 to
7ea5de8
Compare
|
(X-ref : discourse) Thanks for this feature! I have been able to successfully use it on simple models. However, it still fails in more complex examples, such as the following: using DataFrames
data = DataFrame(x = rand(100));
data.y = 1 .+ 2*data.x .+ rand(100);
using GLM
model = lm(@formula(y ~ x + x^2), data) # @formula(y ~ x) works well
using Plots; gr()
using StatsPlots
plot(xlabel="x", ylabel="y", legend=:bottomright)
plot!(data.x, data.y, label="data", seriestype=:scatter)
plot!(model)for which I get the following error / stack trace: julia> plot!(model)
ERROR: DimensionMismatch("second dimension of A, 2, does not match length of x, 3")
Stacktrace:
[1] gemv!(::Array{Float64,1}, ::Char, ::Array{Float64,2}, ::Array{Float64,1}, ::Bo
ol, ::Bool) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.
4/LinearAlgebra/src/matmul.jl:456
[2] mul! at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.4/
LinearAlgebra/src/matmul.jl:66 [inlined]
[3] mul! at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.4/
LinearAlgebra/src/matmul.jl:208 [inlined]
[4] * at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.4/Lin
earAlgebra/src/matmul.jl:47 [inlined]
[5] predict(::LinearModel{GLM.LmResp{Array{Float64,1}},GLM.DensePredChol{Float64,L
inearAlgebra.Cholesky{Float64,Array{Float64,2}}}}, ::Array{Float64,2}; interval::Sy
mbol, level::Float64) at /home/francois/.julia/packages/GLM/6V3fS/src/lm.jl:201
[6] #predict#79 at /home/francois/.julia/packages/StatsModels/pMxlJ/src/statsmodel
.jl:125 [inlined]
[7] macro expansion at /home/francois/.julia/packages/StatsPlots/0rYaj/src/statsmo
dels.jl:4 [inlined]
[8] apply_recipe(::Dict{Symbol,Any}, ::StatsModels.TableRegressionModel{LinearMode
l{GLM.LmResp{Array{Float64,1}},GLM.DensePredChol{Float64,LinearAlgebra.Cholesky{Flo
at64,Array{Float64,2}}}},Array{Float64,2}}) at /home/francois/.julia/packages/Recip
esBase/G4s6f/src/RecipesBase.jl:279
[9] _process_userrecipes(::Plots.Plot{Plots.GRBackend}, ::Dict{Symbol,Any}, ::Tupl
e{StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Array{Float64,1}},GLM.Den
sePredChol{Float64,LinearAlgebra.Cholesky{Float64,Array{Float64,2}}}},Array{Float64
,2}}}) at /home/francois/.julia/packages/Plots/cc8wh/src/pipeline.jl:85
[10] _plot!(::Plots.Plot{Plots.GRBackend}, ::Dict{Symbol,Any}, ::Tuple{StatsModels
.TableRegressionModel{LinearModel{GLM.LmResp{Array{Float64,1}},GLM.DensePredChol{Fl
oat64,LinearAlgebra.Cholesky{Float64,Array{Float64,2}}}},Array{Float64,2}}}) at /ho
me/francois/.julia/packages/Plots/cc8wh/src/plot.jl:178
[11] #plot!#138 at /home/francois/.julia/packages/Plots/cc8wh/src/plot.jl:158 [inl
ined]
[12] plot!(::Plots.Plot{Plots.GRBackend}, ::StatsModels.TableRegressionModel{Linea
rModel{GLM.LmResp{Array{Float64,1}},GLM.DensePredChol{Float64,LinearAlgebra.Cholesk
y{Float64,Array{Float64,2}}}},Array{Float64,2}}) at /home/francois/.julia/packages/
Plots/cc8wh/src/plot.jl:155
[13] plot!(::StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Array{Float64
,1}},GLM.DensePredChol{Float64,LinearAlgebra.Cholesky{Float64,Array{Float64,2}}}},A
rray{Float64,2}}; kw::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{,Tupl
e{}}}) at /home/francois/.julia/packages/Plots/cc8wh/src/plot.jl:150
[14] plot!(::StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Array{Float64
,1}},GLM.DensePredChol{Float64,LinearAlgebra.Cholesky{Float64,Array{Float64,2}}}},A
rray{Float64,2}}) at /home/francois/.julia/packages/Plots/cc8wh/src/plot.jl:144
[15] top-level scope at REPL[32]:1Thanks anyway for StatsPlots, and please don't hesitate to tell me if I can help! |
|
I am getting error |
Here's a basic example usage:
This is really rough and does only address the bivariate case - we're waiting for a general interface to StatsModels (will open an issue there)
cc @Tokazama