OneClassSVM cannot be converted to probabilistic outlier classificator

[IJ1TI]

Describe the bug
Probably Mismatch in internal functions - expected different implementation of scale_minmax

[ Info: Training machine(ProbabilisticUnsupervisedCompositeDetector(normalize = scale_minmax, …), …).
[ Info: Training machine(OneClassSVM(kernel = Polynomial, …), …).
┌ Error: Failed to apply the operation #19.
│ Incoming data:
│ arg of #19 scitype
│ -------------------------------------------
│ Node @959 → OneClassSVM(…) AbstractVector{Continuous}
│
│ Learning network sources:
│ source scitype
│ -------------------------------------------
│ Source @500 Table{AbstractVector{Continuous}}
└ @ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:164
┌ Error: Problem fitting the machine machine(ProbabilisticUnsupervisedCompositeDetector(normalize = scale_minmax, …), …).
└ @ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/machines.jl:695
[ Info: Running type checks...
[ Info: Type checks okay.
ERROR: MethodError: no method matching scale_minmax(::Vector{Float64})
The function scale_minmax exists, but no method is defined for this combination of argument types.

Closest candidates are:
scale_minmax(::Tuple{AbstractVector{<:Real}, AbstractVector{<:Real}})
@ OutlierDetection ~/.julia/packages/OutlierDetection/bjM0k/src/normalization.jl:24

Stacktrace:
[1] (::OutlierDetection.var"#19#22"{OutlierDetection.ProbabilisticUnsupervisedCompositeDetector{…}})(n::Vector{Float64})
@ OutlierDetection ~/.julia/packages/OutlierDetection/bjM0k/src/mlj_wrappers.jl:153
[2] _apply(::Tuple{Node{Nothing, OutlierDetection.var"#19#22"{…}}}; kwargs::@kwargs{rows::Colon})
@ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:153
[3] _apply
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:146 [inlined]
[4] Node
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:143 [inlined]
[5] #77
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:150 [inlined]
[6] map(f::MLJBase.var"#77#78"{@kwargs{rows::Colon}, Tuple{}}, t::Tuple{Node{Nothing, OutlierDetection.var"#19#22"{…}}})
@ Base ./tuple.jl:355
[7] _apply(::Tuple{Node{Nothing, typeof(combine_mean)}}; kwargs::@kwargs{rows::Colon})
@ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:149
[8] _apply
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:146 [inlined]
[9] Node
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:143 [inlined]
[10] #77
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:150 [inlined]
[11] map(f::MLJBase.var"#77#78"{@kwargs{rows::Colon}, Tuple{}}, t::Tuple{Node{Nothing, typeof(combine_mean)}})
@ Base ./tuple.jl:355
[12] _apply(::Tuple{Node{Nothing, typeof(first)}}; kwargs::@kwargs{rows::Colon})
@ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:149
[13] _apply
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:146 [inlined]
[14] Node
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/nodes.jl:143 [inlined]
[15] call_and_copy
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/signatures.jl:81 [inlined]
[16] _broadcast_getindex_evalf
@ ./broadcast.jl:678 [inlined]
[17] _broadcast_getindex
@ ./broadcast.jl:651 [inlined]
[18] #17
@ ./broadcast.jl:1102 [inlined]
[19] ntuple
@ ./ntuple.jl:48 [inlined]
[20] copy
@ ./broadcast.jl:1102 [inlined]
[21] materialize
@ ./broadcast.jl:872 [inlined]
[22] call_and_copy
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/signatures.jl:85 [inlined]
[23] report_supplement
@ ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/signatures.jl:293 [inlined]
[24] report(signature::MLJBase.Signature{@NamedTuple{…}}; supplement::Bool)
@ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/signatures.jl:328
[25] report(signature::MLJBase.Signature{@NamedTuple{predict::Node{…}, transform::Node{…}, report::@NamedTuple{…}}})
@ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/composition/learning_networks/signatures.jl:326
[26] fit(composite::OutlierDetection.ProbabilisticUnsupervisedCompositeDetector{…}, verbosity::Int64, data::Tables.MatrixTable{…})
@ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/composition/models/network_composite.jl:33
[27] fit_only!(mach::Machine{…}; rows::Nothing, verbosity::Int64, force::Bool, composite::Nothing)
@ MLJBase ~/.julia/packages/MLJBase/GY2fM/src/machines.jl:693
[28] fit_only!
@ ~/.julia/packages/MLJBase/GY2fM/src/machines.jl:617 [inlined]
[29] #fit!#66
@ ~/.julia/packages/MLJBase/GY2fM/src/machines.jl:790 [inlined]
[30] fit!
@ ~/.julia/packages/MLJBase/GY2fM/src/machines.jl:787 [inlined]
[31] |>(x::Machine{…}, f::typeof(fit!))
@ Base ./operators.jl:926
[32] top-level scope
@ ~/DevOps/learning_ML/tests/test.jl:23
Some type information was truncated. Use show(err) to see complete types.

To Reproduce
Run the exmaple from documentation or source below

https://juliaai.github.io/MLJ.jl/dev/models/OneClassSVM_LIBSVM/#Outlier-classification-using-a-probability-threshold:

using MLJ
import LIBSVM
import StableRNGs.StableRNG

OneClassSVM = @load OneClassSVM pkg = LIBSVM           ## model type
model = OneClassSVM(kernel=LIBSVM.Kernel.Polynomial) ## instance

rng = StableRNG(123)
Xmatrix = randn(rng, 5, 3)
Xmatrix[1, 1] = 100.0 # outlier 
X = MLJ.table(Xmatrix)

mach = machine(model, X) |> fit!
report(mach).scores

using OutlierDetection
pmodel = ProbabilisticDetector(model) # throws error! 
pmach = machine(pmodel, X) |> fit!

dmodel = BinaryThresholdPredictor(pmodel, threshold=0.9)
dmach = machine(dmodel, X) |> fit!

Versions
Status ~/DevOps/learning_ML/Project.toml
[a93c6f00] DataFrames v1.8.1
[b4f34e82] Distances v0.10.12
[b1bec4e5] LIBSVM v0.8.1
[add582a8] MLJ v0.22.0
[61c7150f] MLJLIBSVMInterface v0.2.2
[6ee0df7b] MLJLinearModels v0.10.1
[1b6a4a23] MLJMultivariateStatsInterface v0.5.4
[5ae90465] MLJScikitLearnInterface v0.7.0
[262411bb] OutlierDetection v0.4.2
[91a5bcdd] Plots v1.41.3
[54e16d92] PrettyPrinting v0.4.2
[ce6b1742] RDatasets v0.8.0
[860ef19b] StableRNGs v1.0.4
[2913bbd2] StatsBase v0.34.9

Julia Version 1.11.5

[ablaom]

Thanks for this, @IJ1TI.

Yes, I'm sorry but the OneClassSVM pkg = LIBSVM supports the API provided by OutlierDetection.jl. The implementation of an MLJ interface for OneClass is pretty old, when there was no formal API for outlier detection. And I don't think there is a SVM detector in the OutlierDetection.jl suite, unless @davnn cares to contradict me.

I'd support changes at MLJLIBSVMInterface to bring OneClassSVM into line, in case you have the expertise and interest, but I don't have the bandwidth to do this myself at present. Nevertheless, I will open an issue there to flag your implicit request.