Large models name change in BetaML

[sylvaticus]

Hello, just to preannounce a large name change in BetaML model names, in order to increase consistency.

These are in master, but not yet on any release, so still time to amend them:

BetaML name	MLJ Interface
PerceptronClassifier	LinearPerceptron
KernelPerceptronClassifier	KernelPerceptron
PegasosClassifier	Pegasos
DecisionTreeEstimator	DecisionTreeClassifier, DecisionTreeRegressor
RandomForestEstimator	RandomForestClassifier, RandomForestRegressor
NeuralNetworkEstimator	NeuralNetworkRegressor, MultitargetNeuralNetworkRegressor, NeuralNetworkClassifier
GMMRegressor1
GMMRegressor2	GaussianMixtureRegressor
KMeansClusterer	KMeans
KMedoidsClusterer	KMedoids
GMMClusterer	GaussianMixtureClusterer
FeatureBasedImputer	SimpleImputer
GMMImputer	GaussianMixtureImputer
RFImputer	RandomForestImputer
UniversalImputer	GeneralImputer
MinMaxScaler
StandardScaler
Scaler
PCA
OneHotEncoder
OrdinalEncoder

Note that the NN models, in their default formulation, do not require autodiff. Also, I am starting to document the various MLJ models in detail using standard struct docstring, if I interpreted the warning message correctly.

[ablaom]

Also, I am starting to document the various MLJ models in detail using standard struct docstring, if I interpreted the warning message correctly.

Yes, you are interpreting this correctly. BTW, our technical writer has gotten quite busy on other packages, so if you could do these yourself, that would be awesome. A couple of notes:

• You need to bump the MLJModelInteface to 1.4, if you are going to auto-generate a header with MLJModelInteface.doc_header (as explained here). In that case the doc string also has to be defined after the model_metadata declaration (or individual trait declarations) instead of immediately above the struct definition.
• You need to remove any descr=... or docstring=... declarations from model_metadata, or remove the docstring(::Type{...}) =... declaration if you have that instead. The trait now fallback to the ordinary docstring.

Tag me in your PR's and I will double check.

[bear-jordan]

Hi @sylvaticus, I am happy to help update docstrings if you need a hand. My email is included on my profile page, so feel free to contact me there

[sylvaticus]

Hi there, I finally released BetaML v0.9 with the new and renamed MLJ interface models. Concerning MLJ interface models documentation, I haven't touched anything not to make confusion with the work of @bear-jordan. Anyhow, new doc can go on v0.9.1 as is non-breaking. @ablaom: let me know if I can already update the page https://alan-turing-institute.github.io/MLJ.jl/dev/list_of_supported_models/ or you need first to do something to account for this new BetaML version)

[bear-jordan]

Hi Antonello,

I will share some progress on the docs this week. Had to travel for work last week, so things got a bit hectic.

Best, Bear

Bear Jordan MS Geology and Data Analytics Georgia Institute of Technology LinkedIn http://www.linkedin.com/in/bear-jordan

On Sun, Oct 2, 2022 at 8:05 AM Antonello Lobianco @.***> wrote:

Hi there, I finally released BetAML v0.9 with the new and renamed MLJ interface models. Concerning MLJ interface models documentation, I haven't touched anything not to make confusion with the work of @bear-jordan https://github.com/bear-jordan. Anyhow, new doc can go on v0.9.1 as is non-breaking. @ablaom https://github.com/ablaom: let me know if I can already update the page https://alan-turing-institute.github.io/MLJ.jl/dev/list_of_supported_models/ or you need first to do something to account for this new BetaML version)

— Reply to this email directly, view it on GitHub https://github.com/alan-turing-institute/MLJ.jl/issues/963#issuecomment-1264627217, or unsubscribe https://github.com/notifications/unsubscribe-auth/AREFPKNK5VJ3R6QRMQYIBT3WBF3BLANCNFSM6AAAAAAQIZD5XI . You are receiving this because you were mentioned.Message ID: @.***>

[ablaom]

@sylvaticus Looks like I need to update the MLJ model registry first. Will get onto that today.

[ablaom]

While doing this I noticed some traces of "MissingImputator" which I'm guessing is an obsolete model from BetaML ?

using MLJModels # after updating registry
julia> models() do m
       m.package_name == "unknown"
       end
1-element Vector{NamedTuple{(:name, :package_name, :is_supervised, :abstract_type, :deep_properties, :docstring, :fit_data_scitype, :human_name, :hyperparameter_ranges, :hyperparameter_types, :hyperparameters, :implemented_methods, :inverse_transform_scitype, :is_pure_julia, :is_wrapper, :iteration_parameter, :load_path, :package_license, :package_url, :package_uuid, :predict_scitype, :prediction_type, :reporting_operations, :reports_feature_importances, :supports_class_weights, :supports_online, :supports_training_losses, :supports_weights, :transform_scitype, :input_scitype, :target_scitype, :output_scitype)}}:
 (name = MissingImputator, package_name = unknown, ... )

MLJModels builds the registry by looking for all subtypes of Model.

Perhaps BetaML has some residual obsolete code somewhere?

[sylvaticus]

Yes, I left it as deprecated with the intention to remove it in BetaML 0.9.

[ablaom]

Okay. Looks like only the package metadata for that model been changed, which probably won't break much, if anything. And your release is tagged breaking anyway.

julia> info("MissingImputator")
(name = "MissingImputator",
 package_name = "unknown",
 is_supervised = false,
 abstract_type = MLJModelInterface.Unsupervised,
 deep_properties = (),
 docstring = "Impute missing values using an Expectation-Maximis...",
 fit_data_scitype =
     Tuple{ScientificTypesBase.Table{<:AbstractVector{<:Union{Missing, ScientificTypesBase.Continuous}}}},
 human_name = "missing imputator",
 hyperparameter_ranges =
     (nothing, nothing, nothing, nothing, nothing, nothing, nothing, nothing, nothing),
 hyperparameter_types = ("Int64",
                         "AbstractVector{Float64}",
                         "Symbol",
                         "Float64",
                         "Float64",
                         "Float64",
                         "String",
                         "BetaML.Api.Verbosity",
                         "Random.AbstractRNG"),
 hyperparameters = (:K,
                    :initial_probmixtures,
                    :mixtures,
                    :tol,
                    :minimum_variance,
                    :minimum_covariance,
                    :initialisation_strategy,
                    :verbosity,
                    :rng),
 implemented_methods = [:fit, :transform],
 inverse_transform_scitype =
     ScientificTypesBase.Table{<:AbstractVector{<:Union{Missing, ScientificTypesBase.Continuous}}},
 is_pure_julia = false,
 is_wrapper = false,
 iteration_parameter = nothing,
 load_path = "BetaML.Imputation.MissingImputator",
 package_license = "unknown",
 package_url = "unknown",
 package_uuid = "unknown",
 predict_scitype = ScientificTypesBase.Unknown,
 prediction_type = :unknown,
 reporting_operations = (),
 reports_feature_importances = false,
 supports_class_weights = false,
 supports_online = false,
 supports_training_losses = false,
 supports_weights = false,
 transform_scitype =
     ScientificTypesBase.Table{<:AbstractVector{<:ScientificTypesBase.Continuous}},
 input_scitype =
     ScientificTypesBase.Table{<:AbstractVector{<:Union{Missing, ScientificTypesBase.Continuous}}},
 target_scitype = ScientificTypesBase.Unknown,
 output_scitype =
     ScientificTypesBase.Table{<:AbstractVector{<:ScientificTypesBase.Continuous}})

[ablaom]

This PR completes the model registry update: https://github.com/JuliaRegistries/General/pull/69552.

Be great to get a PR for the MLJ doc change (list of models)!

[sylvaticus]

Done it: https://github.com/alan-turing-institute/MLJ.jl/pull/968 By the way, I don't know in your snippet why info("MissingImputator") returns that it is NOT pure Julia and the licence is unknown. On the root file of my package I "register" all the MLJ model interfaces as:

function __init__()
    MMI.metadata_pkg.(MLJ_INTERFACED_MODELS,
        name       = "BetaML",
        uuid       = "024491cd-cc6b-443e-8034-08ea7eb7db2b",     # see your Project.toml
        url        = "https://github.com/sylvaticus/BetaML.jl",  # URL to your package repo
        julia      = true,     # is it written entirely in Julia?
        license    = "MIT",    # your package license
        is_wrapper = false,    # does it wrap around some other package?
    )
end

Seems some info is not applied.

[ablaom]

julia> BetaML.MLJ_INTERFACED_MODELS
(LinearPerceptron, KernelPerceptron, Pegasos, DecisionTreeClassifier, DecisionTreeRegressor, RandomForestClassifier, RandomForestRegressor, KMeans, KMedoids, GaussianMixtureClusterer, SimpleImputer, GaussianMixtureImputer, RandomForestImputer, GeneralImputer, NeuralNetworkRegressor, MultitargetNeuralNetworkRegressor, NeuralNetworkClassifier, GaussianMixtureRegressor, MultitargetGaussianMixtureRegressor)

julia> BetaML.MissingImputator in ans
false

[ablaom]

Closed as completed.