SMOTENC requires at least one continuous feature

[ablaom]

"SMOTENC" does not work unless there is at least one continuous feature. At least, that's reported for this python implementation and here we do get a fail for all categoricals:

SMOTENC = @load SMOTENC pkg=Imbalance verbosity=0
balanced_pipe = BalancedModel(pipe, balancer=SMOTENC())

mach = machine(SMOTENC())
transform(mach, X, y)
# ERROR: MethodError: no method matching smotenc(::Matrix{…}, ::CategoricalArrays.CategoricalVector{…}, ::Vector{…}; k::Int64, ratios::Float64, knn_tree::String, rng::Random.TaskLocalRNG)
# The function `smotenc` exists, but no method is defined for this combination of argument types.

# Closest candidates are:
#   smotenc(::Any, ::AbstractVector; k, ratios, knn_tree, rng, try_preserve_type)
#    @ Imbalance ~/.julia/packages/Imbalance/LFVkH/src/oversampling_methods/smotenc/smotenc.jl:315                                                                              
#   smotenc(::AbstractMatrix{<:AbstractFloat}, ::AbstractVector, ::AbstractVector{<:Int64}; k, ratios, knn_tree, rng, try_preserve_type) 
#    @ Imbalance ~/.julia/packages/Imbalance/LFVkH/src/oversampling_methods/smotenc/smotenc.jl:296                                                                              
#   smotenc(::Any, ::Integer; k, ratios, knn_tree, rng, try_preserve_type)
#    @ Imbalance ~/.julia/packages/Imbalance/LFVkH/src/oversampling_methods/smotenc/smotenc.jl:340                                                                              

# Stacktrace:
#  [1] tablify(matrix_func::typeof(Imbalance.smotenc), X::@NamedTuple{…}, y::CategoricalArrays.CategoricalVector{…}; try_preserve_type::Bool, encode_func::typeof(Imbalance.smotenc_encoder), decode_func::typeof(Imbalance.smotenc_decoder), kwargs::@Kwargs{…})
#    @ Imbalance ~/.julia/packages/Imbalance/LFVkH/src/table_wrappers.jl:73
#  [2] tablify
#    @ ~/.julia/packages/Imbalance/LFVkH/src/table_wrappers.jl:57 [inlined]
#  [3] #smotenc#94
#    @ ~/.julia/packages/Imbalance/LFVkH/src/oversampling_methods/smotenc/smotenc.jl:324 [inlined]                                                                                  
#  [4] smotenc
#    @ ~/.julia/packages/Imbalance/LFVkH/src/oversampling_methods/smotenc/smotenc.jl:315 [inlined]                                                                                  
#  [5] transform(s::Imbalance.MLJ.SMOTENC{…}, ::Nothing, X::@NamedTuple{…}, y::CategoricalArrays.CategoricalVector{…})                                                     
#    @ Imbalance.MLJ ~/.julia/packages/Imbalance/LFVkH/src/oversampling_methods/smotenc/interface_mlj.jl:46                                                                         
#  [6] transform(mach::Machine{…}, Xraw::@NamedTuple{…}, Xraw_more::CategoricalArrays.CategoricalVector{…})                                                     
#    @ MLJBase ~/.julia/packages/MLJBase/yVJvJ/src/operations.jl:143
#  [7] top-level scope
#    @ REPL[106]:1
#     Some type information was truncated. Use `show(err)` to see complete types.

@EssamWisam Can you confirm. If so, we should amend the document string.

[EssamWisam]

Yes it's confirmed in the literature and technically makes sense because the method computes an aggregate over continuous features and then uses it for nominal ones.

[EssamWisam]

Actually I have just remembered that I noted this in the docs already:

[EssamWisam]

Shall we close this?

[ablaom]

That's great to have, but the docstring you mention is not exposed in the MLJ interface. The update needs to happen in the MLJ model docstring.