lalvim / partialleastsquaresregressor.jl Goto Github PK

Hi, I have a few questions regarding the outputs of f = fitted_params(mach) and r = report(mach) on a trained mach

• What are each of the objects in the fitted_params(mach) output? After navigating this repo I could find out that the first element is f[1].W, the second element is f[1].b, and the third element is f[1].P; but this is not very clear and definitely not straightforward. It would be nice to have a description of how to access these objects and what they are in the docs of this package -there is a lot of inconsistency in terminology out there, and it is not easy to know what they actually are.
• Is report(mach) expected to return nothing?
• What is the best way to report feature importance with the matrices available (W, b, P)? Alternatively, it would be nice to have some metric of feature importance after fitting a model. (https://learnche.org/pid/latent-variable-modelling/projection-to-latent-structures/coefficient-plots-in-pls)

Is the 'model.P' attribute from PartialLeastSquaresRegressor.jl equivalent to the x_rotations_ attribute from SKLearn?

The SKLearn documentation describes this attribute as: "The projection matrix used to transform X"

Follow the approach given: Martens H., NÊs T. Multivariate Calibration. Wiley: New York, 1989.
as shown here

https://github.com/caseykneale/ChemometricsTools.jl/blob/d8cd288ae76b221274a54cc204cd146791bddf98/src/RegressionModels.jl#L212

This turns inference into a single matmul, because PLS does truly follow Y = XB when center scaled. Should work for PLS1 and PLS2.

The example 1 in the readme has issues:

julia> regressor = PLSRegressor(n_factors=2)
ERROR: UndefVarError: PLSRegressor not defined

This is easily fixed by adding
using PartialLeastSquaresRegressor: PLSRegressor
but maybe you really should export that type from the package?

Then a bit later this happens:

julia> pls_model = @pipeline Standardizer regressor target=Standardizer

ERROR: LoadError: The `@pipeline` macro is deprecated. For pipelines without target transformations use pipe syntax, as in `ContinuousEncoder() |> Standardizer() |> my_classifier`. For details and advanced optioins, query the `Pipeline` docstring. To wrap a supervised model in a target transformation, use `TransformedTargetModel`, as in `TransformedTargetModel(my_regressor, target=Standardizer())`
in expression starting at REPL[16]:1

We would love to use the package, but we need a working example.

In the long term, I recommend using Literate.jl, to show working examples because then they are tested as part of CI, whereas examples in a README are not. But in the short run could you please fix the readme? Here is one Literate example:
https://jefffessler.github.io/ScoreMatching.jl/dev/generated/examples/01-overview/

julia> Pkg.add("PLSRegressor")
Resolving package versions...
ERROR: Unsatisfiable requirements detected for package PLSRegressor [fba1ee03]:
PLSRegressor [fba1ee03] log:
├─possible versions are: 1.0.1 or uninstalled
├─restricted to versions * by an explicit requirement, leaving only versions 1.0.1
└─restricted by julia compatibility requirements to versions: uninstalled — no versions left
Stacktrace:
[1] #propagate_constraints!#61(::Bool, ::Function, ::Pkg.GraphType.Graph, ::Set{Int64}) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/GraphType.jl:1007
[2] propagate_constraints! at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/GraphType.jl:948 [inlined]
[3] #simplify_graph!#121(::Bool, ::Function, ::Pkg.GraphType.Graph, ::Set{Int64}) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/GraphType.jl:1462
[4] simplify_graph! at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/GraphType.jl:1462 [inlined] (repeats 2 times)
[5] resolve_versions!(::Pkg.Types.Context, ::Array{Pkg.Types.PackageSpec,1}, ::Nothing) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/Operations.jl:371
[6] resolve_versions! at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/Operations.jl:315 [inlined]
[7] #add_or_develop#63(::Array{Base.UUID,1}, ::Symbol, ::Function, ::Pkg.Types.Context, ::Array{Pkg.Types.PackageSpec,1}) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/Operations.jl:1172
[8] #add_or_develop at ./none:0 [inlined]
[9] #add_or_develop#17(::Symbol, ::Bool, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::Function, ::Pkg.Types.Context, ::Array{Pkg.Types.PackageSpec,1}) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/API.jl:59
[10] #add_or_develop at ./none:0 [inlined]
[11] #add_or_develop#16 at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/API.jl:36 [inlined]
[12] #add_or_develop at ./none:0 [inlined]
[13] #add_or_develop#13 at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/API.jl:34 [inlined]
[14] #add_or_develop at ./none:0 [inlined]
[15] #add_or_develop#12(::Base.Iterators.Pairs{Symbol,Symbol,Tuple{Symbol},NamedTuple{(:mode,),Tuple{Symbol}}}, ::Function, ::String) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/API.jl:33
[16] #add_or_develop at ./none:0 [inlined]
[17] #add#22 at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/API.jl:64 [inlined]
[18] add(::String) at /Users/julia/buildbot/worker/package_macos64/build/usr/share/julia/stdlib/v1.1/Pkg/src/API.jl:64
[19] top-level scope at none:0

How can loadings and scores be extracted after performing fit using fit!(pls_machine, rows = train)?

A Fast PLS version
https://www.rdocumentation.org/packages/plsdepot/versions/0.1.17/topics/simpls

S. de Jong. SIMPLS: An Alternative approach to partial least squares regression. Chemometrics and Intelligent Laboratory Systems, 18, 1993 (251-263).

Add diagnostics like Leverage, Explained variance in X&Y, and Q& Hotelling statistics as in here:

https://github.com/caseykneale/ChemometricsTools.jl/blob/master/src/ModelAnalysis.jl#L21

As described in https://discourse.julialang.org/t/ann-plans-for-removing-packages-that-do-not-yet-support-1-0-from-the-general-registry/ we are planning on removing packages that do not support 1.0 from the General registry. This package has been detected to not support 1.0 and is thus slated to be removed. The removal of packages from the registry will happen approximately a month after this issue is open.

To transition to the new Pkg system using Project.toml, see https://github.com/JuliaRegistries/Registrator.jl#transitioning-from-require-to-projecttoml.
To then tag a new version of the package, see https://github.com/JuliaRegistries/Registrator.jl#via-the-github-app.

If you believe this package has erroneously been detected as not supporting 1.0 or have any other questions, don't hesitate to discuss it here or in the thread linked at the top of this post.

A N x n_factors matrix, P, is allocated on line 59 in kpls.jl and then populated on line 103 but is otherwise unused. Not sure if julia optimizes it away, but if not it can be a pretty sizeable allocation that should be removed.

Hi @lalvim
I'm excited about your updated package!
Have you considered making an announcement here: https://discourse.julialang.org/c/community/packages/47
It can get your package some attention.

I tried to fit uninformative data (random, uniform, and centered) with PLS2 and the regressor was unable to learn the baseline (note that I am using the MLJ interface from #10).

regressor = PLS(n_factors=1)

X = rand(1000, 5) .- 0.5
y = rand(1000, 2) .- 0.5
plsmachine = MLJ.machine(regressor, MLJ.table(X), MLJ.table(y))
MLJ.fit!(plsmachine)

pred = MLJ.predict(plsmachine)
yhat = MLJ.matrix(pred)

# Error of the model
println(sum((y .- yhat).^2))  # 249.26
# Baseline prediction yhat = 0
println(sum(y.^2))  # 166.33

I would expect the error to be not worse for the PLS2 model here since by learning every internal parameters to be zero, it would always return [0, 0] as output and match the baseline prediction.

Scikit learn version on the other hand works as expected. It doesn't quite learn all parameters to be zero, but the final error matches the baseline's one.

Hi and thank you for this package!
have you considered porting it to MLJ.jl?

For some data sets training is failing. Given the MethodError thrown, this looks like a bug to me:

julia> using MLJBase, PartialLeastSquaresRegressor

julia> X, y = @load_boston;

julia> machine(PartialLeastSquaresRegressor.PLSRegressor(), X, y) |> fit!
[ Info: Training machine(PLSRegressor(n_factors = 1), …).
┌ Error: Problem fitting the machine machine(PLSRegressor(n_factors = 1), …). 
└ @ MLJBase ~/.julia/packages/MLJBase/wnJff/src/machines.jl:617
[ Info: Running type checks... 
[ Info: Type checks okay. 
ERROR: MethodError: no method matching check_constant_cols(::SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true})
Closest candidates are:
  check_constant_cols(::Matrix{T}) where T<:AbstractFloat at /Users/anthony/.julia/packages/PartialLeastSquaresRegressor/OrIoJ/src/utils.jl:26
  check_constant_cols(::Vector{T}) where T<:AbstractFloat at /Users/anthony/.julia/packages/PartialLeastSquaresRegressor/OrIoJ/src/utils.jl:27
Stacktrace:
 [1] fit(m::PartialLeastSquaresRegressor.PLSRegressor, verbosity::Int64, X::NamedTuple{(:Crim, :Zn, :Indus, :NOx, :Rm, :Age, :Dis, :Rad, :Tax, :PTRatio, :Black, :LStat), NTuple{12, SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}}, Y::SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true})
   @ PartialLeastSquaresRegressor ~/.julia/packages/PartialLeastSquaresRegressor/OrIoJ/src/mlj_interface.jl:65
 [2] fit_only!(mach::Machine{PartialLeastSquaresRegressor.PLSRegressor, true}; rows::Nothing, verbosity::Int64, force::Bool)
   @ MLJBase ~/.julia/packages/MLJBase/wnJff/src/machines.jl:615
 [3] fit_only!
   @ ~/.julia/packages/MLJBase/wnJff/src/machines.jl:568 [inlined]
 [4] #fit!#52
   @ ~/.julia/packages/MLJBase/wnJff/src/machines.jl:683 [inlined]
 [5] fit!
   @ ~/.julia/packages/MLJBase/wnJff/src/machines.jl:681 [inlined]
 [6] |>(x::Machine{PartialLeastSquaresRegressor.PLSRegressor, true}, f::typeof(fit!))
   @ Base ./operators.jl:858
 [7] top-level scope
   @ REPL[162]:1
 [8] top-level scope
   @ ~/.julia/packages/CUDA/fAEDi/src/initialization.jl:52

It would be nice a good documentation.

The tag name "v1.0" is not of the appropriate SemVer form (vX.Y.Z).
cc: @lalvim