Common MLJ Workflows

Data ingestion

import RDatasets
channing = RDatasets.dataset("boot", "channing")

julia> first(channing, 4)
4×5 DataFrame
 Row │ Sex   Entry  Exit   Time   Cens
     │ Cat…  Int32  Int32  Int32  Int32
─────┼──────────────────────────────────
   1 │ Male    782    909    127      1
   2 │ Male   1020   1128    108      1
   3 │ Male    856    969    113      1
   4 │ Male    915    957     42      1

Inspecting metadata, including column scientific types:

schema(channing)

┌─────────┬──────────────────────────────────┬───────────────┐
│ _.names │ _.types                          │ _.scitypes    │
├─────────┼──────────────────────────────────┼───────────────┤
│ Sex     │ CategoricalValue{String, UInt32} │ Multiclass{2} │
│ Entry   │ Int64                            │ Count         │
│ Exit    │ Int64                            │ Count         │
│ Time    │ Int64                            │ Count         │
│ Cens    │ Int64                            │ Count         │
└─────────┴──────────────────────────────────┴───────────────┘
_.nrows = 462

Unpacking data and shuffling rows:

y, X =  unpack(channing,
               ==(:Exit),            # y is the :Exit column
               !=(:Time));           # X is the rest, except :Time

Note: Before julia 1.2, replace !=(:Time) with col -> col != :Time.

julia> y[1:4]
4-element Vector{Int32}:
  909
 1128
  969
  957

Fixing wrong scientfic types in X:

X = coerce(X, :Exit=>Continuous, :Entry=>Continuous, :Cens=>Multiclass)

julia> schema(X)
┌─────────┬─────────────────────────────────┬───────────────┐
│ _.names │ _.types                         │ _.scitypes    │
├─────────┼─────────────────────────────────┼───────────────┤
│ Sex     │ CategoricalValue{String, UInt8} │ Multiclass{2} │
│ Entry   │ Float64                         │ Continuous    │
│ Cens    │ CategoricalValue{Int32, UInt32} │ Multiclass{2} │
└─────────┴─────────────────────────────────┴───────────────┘
_.nrows = 462

Loading a built-in supervised dataset:

table = load_iris()
schema(table)

┌──────────────┬──────────────────────────────────┬───────────────┐
│ _.names      │ _.types                          │ _.scitypes    │
├──────────────┼──────────────────────────────────┼───────────────┤
│ sepal_length │ Float64                          │ Continuous    │
│ sepal_width  │ Float64                          │ Continuous    │
│ petal_length │ Float64                          │ Continuous    │
│ petal_width  │ Float64                          │ Continuous    │
│ target       │ CategoricalValue{String, UInt32} │ Multiclass{3} │
└──────────────┴──────────────────────────────────┴───────────────┘
_.nrows = 150

Loading a built-in data set already split into X and y:

X, y = @load_iris;
selectrows(X, 1:4) # selectrows works for any Tables.jl table

(sepal_length = [5.1, 4.9, 4.7, 4.6],
 sepal_width = [3.5, 3.0, 3.2, 3.1],
 petal_length = [1.4, 1.4, 1.3, 1.5],
 petal_width = [0.2, 0.2, 0.2, 0.2],)

y[1:4]

4-element CategoricalArray{String,1,UInt32}:
 "setosa"
 "setosa"
 "setosa"
 "setosa"

Model search

Reference: Model Search

Searching for a supervised model:

X, y = @load_boston
models(matching(X, y))

59-element Vector{NamedTuple{(:name, :package_name, :is_supervised, :abstract_type, :deep_properties, :docstring, :fit_data_scitype, :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, :supports_class_weights, :supports_online, :supports_training_losses, :supports_weights, :transform_scitype, :input_scitype, :target_scitype, :output_scitype), T} where T<:Tuple}:
 (name = ARDRegressor, package_name = ScikitLearn, ... )
 (name = AdaBoostRegressor, package_name = ScikitLearn, ... )
 (name = BaggingRegressor, package_name = ScikitLearn, ... )
 (name = BayesianRidgeRegressor, package_name = ScikitLearn, ... )
 (name = ConstantRegressor, package_name = MLJModels, ... )
 (name = DecisionTreeRegressor, package_name = BetaML, ... )
 (name = DecisionTreeRegressor, package_name = DecisionTree, ... )
 (name = DeterministicConstantRegressor, package_name = MLJModels, ... )
 (name = DummyRegressor, package_name = ScikitLearn, ... )
 (name = ElasticNetCVRegressor, package_name = ScikitLearn, ... )
 ⋮
 (name = RidgeRegressor, package_name = MultivariateStats, ... )
 (name = RidgeRegressor, package_name = ScikitLearn, ... )
 (name = RobustRegressor, package_name = MLJLinearModels, ... )
 (name = SGDRegressor, package_name = ScikitLearn, ... )
 (name = SVMLinearRegressor, package_name = ScikitLearn, ... )
 (name = SVMNuRegressor, package_name = ScikitLearn, ... )
 (name = SVMRegressor, package_name = ScikitLearn, ... )
 (name = TheilSenRegressor, package_name = ScikitLearn, ... )
 (name = XGBoostRegressor, package_name = XGBoost, ... )

models(matching(X, y))[6]

A simple Decision Tree for regression with support for Missing data, from the Beta Machine Learning Toolkit (BetaML).
→ based on [BetaML](https://github.com/sylvaticus/BetaML.jl).
→ do `@load DecisionTreeRegressor pkg="BetaML"` to use the model.
→ do `?DecisionTreeRegressor` for documentation.
(name = "DecisionTreeRegressor",
 package_name = "BetaML",
 is_supervised = true,
 abstract_type = Deterministic,
 deep_properties = (),
 docstring = "A simple Decision Tree for regression with support for Missing data, from the Beta Machine Learning Toolkit (BetaML).\n→ based on [BetaML](https://github.com/sylvaticus/BetaML.jl).\n→ do `@load DecisionTreeRegressor pkg=\"BetaML\"` to use the model.\n→ do `?DecisionTreeRegressor` for documentation.",
 fit_data_scitype = Tuple{Table{_s48} where _s48<:Union{AbstractVector{_s47} where _s47<:Missing, AbstractVector{_s47} where _s47<:Known}, AbstractVector{_s4854} where _s4854<:Continuous},
 hyperparameter_ranges = (nothing, nothing, nothing, nothing, nothing, nothing),
 hyperparameter_types = ("Int64", "Float64", "Int64", "Int64", "Function", "Random.AbstractRNG"),
 hyperparameters = (:maxDepth, :minGain, :minRecords, :maxFeatures, :splittingCriterion, :rng),
 implemented_methods = [:fit, :predict],
 inverse_transform_scitype = Unknown,
 is_pure_julia = true,
 is_wrapper = false,
 iteration_parameter = nothing,
 load_path = "BetaML.Trees.DecisionTreeRegressor",
 package_license = "MIT",
 package_url = "https://github.com/sylvaticus/BetaML.jl",
 package_uuid = "024491cd-cc6b-443e-8034-08ea7eb7db2b",
 predict_scitype = AbstractVector{_s4854} where _s4854<:Continuous,
 prediction_type = :deterministic,
 supports_class_weights = false,
 supports_online = false,
 supports_training_losses = false,
 supports_weights = false,
 transform_scitype = Unknown,
 input_scitype = Table{_s48} where _s48<:Union{AbstractVector{_s47} where _s47<:Missing, AbstractVector{_s47} where _s47<:Known},
 target_scitype = AbstractVector{_s4854} where _s4854<:Continuous,
 output_scitype = Unknown,)

Instantiating a model

Reference: Getting Started, Loading Model Code

Tree = @load DecisionTreeClassifier pkg=DecisionTree
tree = Tree(min_samples_split=5, max_depth=4)

DecisionTreeClassifier(
    max_depth = 4,
    min_samples_leaf = 1,
    min_samples_split = 5,
    min_purity_increase = 0.0,
    n_subfeatures = 0,
    post_prune = false,
    merge_purity_threshold = 1.0,
    pdf_smoothing = 0.0,
    display_depth = 5,
    rng = Random._GLOBAL_RNG()) @396

tree = (@load DecisionTreeClassifier)()
tree.min_samples_split = 5
tree.max_depth = 4

Evaluating a model

Reference: Evaluating Model Performance

X, y = @load_boston
KNN = @load KNNRegressor
knn = KNN()
evaluate(knn, X, y, resampling=CV(nfolds=5), measure=[RootMeanSquaredError(), MeanAbsoluteError()])

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────┬─────────────┬───────────┬───────────────────────
│ measure                     │ measurement │ operation │ per_fold             ⋯
├─────────────────────────────┼─────────────┼───────────┼───────────────────────
│ RootMeanSquaredError() @852 │ 8.77        │ predict   │ [8.53, 8.8, 10.7, 9. ⋯
│ MeanAbsoluteError() @788    │ 6.02        │ predict   │ [6.52, 5.7, 7.65, 6. ⋯
└─────────────────────────────┴─────────────┴───────────┴───────────────────────
                                                                1 column omitted

Note RootMeanSquaredError() has alias rms and MeanAbsoluteError() has alias mae.

Do measures() to list all losses and scores and their aliases.

Basic fit/evaluate/predict by hand:

Reference: Getting Started, Machines, Evaluating Model Performance, Performance Measures

crabs = load_crabs() |> DataFrames.DataFrame
schema(crabs)

┌─────────┬──────────────────────────────────┬───────────────┐
│ _.names │ _.types                          │ _.scitypes    │
├─────────┼──────────────────────────────────┼───────────────┤
│ sp      │ CategoricalValue{String, UInt32} │ Multiclass{2} │
│ sex     │ CategoricalValue{String, UInt32} │ Multiclass{2} │
│ index   │ Int64                            │ Count         │
│ FL      │ Float64                          │ Continuous    │
│ RW      │ Float64                          │ Continuous    │
│ CL      │ Float64                          │ Continuous    │
│ CW      │ Float64                          │ Continuous    │
│ BD      │ Float64                          │ Continuous    │
└─────────┴──────────────────────────────────┴───────────────┘
_.nrows = 200

y, X = unpack(crabs, ==(:sp), !in([:index, :sex]); rng=123)


Tree = @load DecisionTreeClassifier pkg=DecisionTree

DecisionTreeClassifier(
    max_depth = 2,
    min_samples_leaf = 1,
    min_samples_split = 2,
    min_purity_increase = 0.0,
    n_subfeatures = 0,
    post_prune = false,
    merge_purity_threshold = 1.0,
    pdf_smoothing = 0.0,
    display_depth = 5,
    rng = Random._GLOBAL_RNG()) @844

Bind the model and data together in a machine , which will additionally store the learned parameters (fitresults) when fit:

mach = machine(tree, X, y)

Machine{DecisionTreeClassifier,…} @133 trained 0 times; caches data
  args: 
    1:	Source @711 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @581 ⏎ `AbstractVector{Multiclass{2}}`

Split row indices into training and evaluation rows:

train, test = partition(eachindex(y), 0.7); # 70:30 split

([1, 2, 3, 4, 5, 6, 7, 8, 9, 10  …  131, 132, 133, 134, 135, 136, 137, 138, 139, 140], [141, 142, 143, 144, 145, 146, 147, 148, 149, 150  …  191, 192, 193, 194, 195, 196, 197, 198, 199, 200])

Fit on train and evaluate on test:

fit!(mach, rows=train)
yhat = predict(mach, X[test,:])
mean(LogLoss(tol=1e-4)(yhat, y[test]))

1.0788055664326648

Note LogLoss() has aliases log_loss and cross_entropy.

Run measures() to list all losses and scores and their aliases ("instances").

Predict on new data:

Xnew = (FL = rand(3), RW = rand(3), CL = rand(3), CW = rand(3), BD =rand(3))
predict(mach, Xnew)      # a vector of distributions

3-element MLJBase.UnivariateFiniteVector{Multiclass{2}, String, UInt32, Float64}:
 UnivariateFinite{Multiclass{2}}(B=>0.667, O=>0.333)
 UnivariateFinite{Multiclass{2}}(B=>0.667, O=>0.333)
 UnivariateFinite{Multiclass{2}}(B=>0.667, O=>0.333)

predict_mode(mach, Xnew) # a vector of point-predictions

3-element CategoricalArray{String,1,UInt32}:
 "B"
 "B"
 "B"

More performance evaluation examples

Evaluating model + data directly:

evaluate(tree, X, y,
         resampling=Holdout(fraction_train=0.7, shuffle=true, rng=1234),
         measure=[LogLoss(), Accuracy()])

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────────┬─────────────┬──────────────┬──────────┐
│ measure                         │ measurement │ operation    │ per_fold │
├─────────────────────────────────┼─────────────┼──────────────┼──────────┤
│ LogLoss(tol = 2.22045e-16) @449 │ 1.12        │ predict      │ [1.12]   │
│ Accuracy() @377                 │ 0.683       │ predict_mode │ [0.683]  │
└─────────────────────────────────┴─────────────┴──────────────┴──────────┘

If a machine is already defined, as above:

evaluate!(mach,
          resampling=Holdout(fraction_train=0.7, shuffle=true, rng=1234),
          measure=[LogLoss(), Accuracy()])

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────────┬─────────────┬──────────────┬──────────┐
│ measure                         │ measurement │ operation    │ per_fold │
├─────────────────────────────────┼─────────────┼──────────────┼──────────┤
│ LogLoss(tol = 2.22045e-16) @449 │ 1.12        │ predict      │ [1.12]   │
│ Accuracy() @377                 │ 0.683       │ predict_mode │ [0.683]  │
└─────────────────────────────────┴─────────────┴──────────────┴──────────┘

Using cross-validation:

evaluate!(mach, resampling=CV(nfolds=5, shuffle=true, rng=1234),
          measure=[LogLoss(), Accuracy()])

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────────┬─────────────┬──────────────┬────────────────
│ measure                         │ measurement │ operation    │ per_fold      ⋯
├─────────────────────────────────┼─────────────┼──────────────┼────────────────
│ LogLoss(tol = 2.22045e-16) @449 │ 0.748       │ predict      │ [0.552, 0.534 ⋯
│ Accuracy() @377                 │ 0.7         │ predict_mode │ [0.775, 0.7,  ⋯
└─────────────────────────────────┴─────────────┴──────────────┴────────────────
                                                                1 column omitted

With user-specified train/test pairs of row indices:

f1, f2, f3 = 1:13, 14:26, 27:36
pairs = [(f1, vcat(f2, f3)), (f2, vcat(f3, f1)), (f3, vcat(f1, f2))];
evaluate!(mach,
          resampling=pairs,
          measure=[LogLoss(), Accuracy()])

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────────┬─────────────┬──────────────┬────────────────
│ measure                         │ measurement │ operation    │ per_fold      ⋯
├─────────────────────────────────┼─────────────┼──────────────┼────────────────
│ LogLoss(tol = 2.22045e-16) @449 │ 4.36        │ predict      │ [5.1, 4.97, 3 ⋯
│ Accuracy() @377                 │ 0.735       │ predict_mode │ [0.696, 0.739 ⋯
└─────────────────────────────────┴─────────────┴──────────────┴────────────────
                                                                1 column omitted

Changing a hyperparameter and re-evaluating:

tree.max_depth = 3
evaluate!(mach,
          resampling=CV(nfolds=5, shuffle=true, rng=1234),
          measure=[LogLoss(), Accuracy()])

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────────┬─────────────┬──────────────┬────────────────
│ measure                         │ measurement │ operation    │ per_fold      ⋯
├─────────────────────────────────┼─────────────┼──────────────┼────────────────
│ LogLoss(tol = 2.22045e-16) @449 │ 1.19        │ predict      │ [1.26, 0.2, 0 ⋯
│ Accuracy() @377                 │ 0.865       │ predict_mode │ [0.8, 0.95, 0 ⋯
└─────────────────────────────────┴─────────────┴──────────────┴────────────────
                                                                1 column omitted

Inspecting training results

Fit a ordinary least square model to some synthetic data:

x1 = rand(100)
x2 = rand(100)

X = (x1=x1, x2=x2)
y = x1 - 2x2 + 0.1*rand(100);

OLS = @load LinearRegressor pkg=GLM
ols = OLS()
mach =  machine(ols, X, y) |> fit!

Machine{LinearRegressor,…} @472 trained 1 time; caches data
  args: 
    1:	Source @945 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @160 ⏎ `AbstractVector{Continuous}`

Get a named tuple representing the learned parameters, human-readable if appropriate:

fitted_params(mach)

(coef = [0.9959827498805456, -2.003470349844367],
 intercept = 0.04967249818365997,)

Get other training-related information:

report(mach)

(deviance = 0.07719711815810197,
 dof_residual = 97.0,
 stderror = [0.00987402731271781, 0.010175807818976984, 0.007617698923428437],
 vcov = [9.749641537229731e-5 3.4325182211293386e-6 -5.036761994370504e-5; 3.4325182211293386e-6 0.00010354706476875311 -5.1647592722261926e-5; -5.036761994370504e-5 -5.1647592722261926e-5 5.8029336888002774e-5],)

Basic fit/transform for unsupervised models

Load data:

X, y = @load_iris
train, test = partition(eachindex(y), 0.97, shuffle=true, rng=123)

([125, 100, 130, 9, 70, 148, 39, 64, 6, 107  …  110, 59, 139, 21, 112, 144, 140, 72, 109, 41], [106, 147, 47, 5])

Instantiate and fit the model/machine:

PCA = @load PCA
pca = PCA(maxoutdim=2)
mach = machine(pca, X)
fit!(mach, rows=train)

Machine{PCA,…} @353 trained 1 time; caches data
  args: 
    1:	Source @700 ⏎ `Table{AbstractVector{Continuous}}`

Transform selected data bound to the machine:

transform(mach, rows=test);

(x1 = [-3.3942826854483243, -1.5219827578765068, 2.538247455185219, 2.7299639893931373],
 x2 = [0.5472450223745241, -0.36842368617126214, 0.5199299511335698, 0.3448466122232363],)

Transform new data:

Xnew = (sepal_length=rand(3), sepal_width=rand(3),
        petal_length=rand(3), petal_width=rand(3));
transform(mach, Xnew)

(x1 = [4.5854266496506275, 4.680104921343005, 4.614666635426547],
 x2 = [-4.467800728319477, -4.294059428101639, -4.599969892097004],)

Inverting learned transformations

y = rand(100);
stand = Standardizer()
mach = machine(stand, y)
fit!(mach)
z = transform(mach, y);
@assert inverse_transform(mach, z) ≈ y # true

[ Info: Training Machine{Standardizer,…} @715.

Nested hyperparameter tuning

Reference: Tuning Models

Define a model with nested hyperparameters:

Tree = @load DecisionTreeClassifier pkg=DecisionTree
tree = Tree()
forest = EnsembleModel(atom=tree, n=300)

ProbabilisticEnsembleModel(
    atom = DecisionTreeClassifier(
            max_depth = -1,
            min_samples_leaf = 1,
            min_samples_split = 2,
            min_purity_increase = 0.0,
            n_subfeatures = 0,
            post_prune = false,
            merge_purity_threshold = 1.0,
            pdf_smoothing = 0.0,
            display_depth = 5,
            rng = Random._GLOBAL_RNG()),
    atomic_weights = Float64[],
    bagging_fraction = 0.8,
    rng = Random._GLOBAL_RNG(),
    n = 300,
    acceleration = CPU1{Nothing}(nothing),
    out_of_bag_measure = Any[]) @073

Define ranges for hyperparameters to be tuned:

r1 = range(forest, :bagging_fraction, lower=0.5, upper=1.0, scale=:log10)

typename(MLJBase.NumericRange)(Float64, :bagging_fraction, ... )

r2 = range(forest, :(atom.n_subfeatures), lower=1, upper=4) # nested

typename(MLJBase.NumericRange)(Int64, :(atom.n_subfeatures), ... )

Wrap the model in a tuning strategy:

tuned_forest = TunedModel(model=forest,
                          tuning=Grid(resolution=12),
                          resampling=CV(nfolds=6),
                          ranges=[r1, r2],
                          measure=BrierScore())

ProbabilisticTunedModel(
    model = ProbabilisticEnsembleModel(
            atom = DecisionTreeClassifier @490,
            atomic_weights = Float64[],
            bagging_fraction = 0.8,
            rng = Random._GLOBAL_RNG(),
            n = 300,
            acceleration = CPU1{Nothing}(nothing),
            out_of_bag_measure = Any[]),
    tuning = Grid(
            goal = nothing,
            resolution = 12,
            shuffle = true,
            rng = Random._GLOBAL_RNG()),
    resampling = CV(
            nfolds = 6,
            shuffle = false,
            rng = Random._GLOBAL_RNG()),
    measure = BrierScore(),
    weights = nothing,
    operation = nothing,
    range = MLJBase.NumericRange{T, MLJBase.Bounded, Symbol} where T[NumericRange{Float64,…} @234, NumericRange{Int64,…} @751],
    selection_heuristic = MLJTuning.NaiveSelection(nothing),
    train_best = true,
    repeats = 1,
    n = nothing,
    acceleration = CPU1{Nothing}(nothing),
    acceleration_resampling = CPU1{Nothing}(nothing),
    check_measure = true,
    cache = true) @146

Bound the wrapped model to data:

mach = machine(tuned_forest, X, y)

Machine{ProbabilisticTunedModel{Grid,…},…} @813 trained 0 times; caches data
  args: 
    1:	Source @813 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @646 ⏎ `AbstractVector{Multiclass{3}}`

Fitting the resultant machine optimizes the hyperparameters specified in range, using the specified tuning and resampling strategies and performance measure (possibly a vector of measures), and retrains on all data bound to the machine:

fit!(mach)

Machine{ProbabilisticTunedModel{Grid,…},…} @813 trained 1 time; caches data
  args: 
    1:	Source @813 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @646 ⏎ `AbstractVector{Multiclass{3}}`

Inspecting the optimal model:

F = fitted_params(mach)

(best_model = ProbabilisticEnsembleModel{DecisionTreeClassifier} @631,
 best_fitted_params = (fitresult = WrappedEnsemble{Tuple{Node{Float64,…},…},…} @797,),)

F.best_model

ProbabilisticEnsembleModel(
    atom = DecisionTreeClassifier(
            max_depth = -1,
            min_samples_leaf = 1,
            min_samples_split = 2,
            min_purity_increase = 0.0,
            n_subfeatures = 3,
            post_prune = false,
            merge_purity_threshold = 1.0,
            pdf_smoothing = 0.0,
            display_depth = 5,
            rng = Random._GLOBAL_RNG()),
    atomic_weights = Float64[],
    bagging_fraction = 0.5325205447199813,
    rng = Random._GLOBAL_RNG(),
    n = 300,
    acceleration = CPU1{Nothing}(nothing),
    out_of_bag_measure = Any[]) @631

Inspecting details of tuning procedure:

r = report(mach);
keys(r)

(:best_model, :best_history_entry, :history, :best_report, :plotting)

r.history[[1,end]]

2-element Vector{NamedTuple{(:model, :measure, :measurement, :per_fold), Tuple{MLJEnsembles.ProbabilisticEnsembleModel{MLJDecisionTreeInterface.DecisionTreeClassifier}, Vector{BrierScore}, Vector{Float64}, Vector{Vector{Float64}}}}}:
 (model = ProbabilisticEnsembleModel{DecisionTreeClassifier} @561, measure = [BrierScore() @840], measurement = [-0.11679733333333316], per_fold = [[-0.009289777777777886, -0.00014666666666668604, -0.1542799999999998, -0.15817777777777736, -0.15875377777777747, -0.22013599999999975]])
 (model = ProbabilisticEnsembleModel{DecisionTreeClassifier} @210, measure = [BrierScore() @840], measurement = [-0.11751856378600833], per_fold = [[-0.028611160493827253, -0.008150222222222387, -0.18381303703703725, -0.1420677777777778, -0.1652894074074074, -0.17717977777777796]])

Visualizing these results:

using Plots
plot(mach)

Predicting on new data using the optimized model:

predict(mach, Xnew)

3-element Vector{UnivariateFinite{Multiclass{3}, String, UInt32, Float64}}:
 UnivariateFinite{Multiclass{3}}(versicolor=>0.0, virginica=>0.0, setosa=>1.0)
 UnivariateFinite{Multiclass{3}}(versicolor=>0.113, virginica=>0.00333, setosa=>0.883)
 UnivariateFinite{Multiclass{3}}(versicolor=>0.0, virginica=>0.0, setosa=>1.0)

Constructing a linear pipeline

Reference: Composing Models

Constructing a linear (unbranching) pipeline with a learned target transformation/inverse transformation:

X, y = @load_reduced_ames
KNN = @load KNNRegressor
pipe = @pipeline(X -> coerce(X, :age=>Continuous),
                 OneHotEncoder,
                 KNN(K=3),
                 target = Standardizer)

Pipeline264(
    one_hot_encoder = OneHotEncoder(
            features = Symbol[],
            drop_last = false,
            ordered_factor = true,
            ignore = false),
    knn_regressor = KNNRegressor(
            K = 3,
            algorithm = :kdtree,
            metric = Distances.Euclidean(0.0),
            leafsize = 10,
            reorder = true,
            weights = NearestNeighborModels.Uniform()),
    target = Standardizer(
            features = Symbol[],
            ignore = false,
            ordered_factor = false,
            count = false)) @151

Evaluating the pipeline (just as you would any other model):

pipe.knn_regressor.K = 2
pipe.one_hot_encoder.drop_last = true
evaluate(pipe, X, y, resampling=Holdout(), measure=RootMeanSquaredError(), verbosity=2)

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────┬─────────────┬───────────┬───────────┐
│ measure                     │ measurement │ operation │ per_fold  │
├─────────────────────────────┼─────────────┼───────────┼───────────┤
│ RootMeanSquaredError() @852 │ 53100.0     │ predict   │ [53100.0] │
└─────────────────────────────┴─────────────┴───────────┴───────────┘

Inspecting the learned parameters in a pipeline:

mach = machine(pipe, X, y) |> fit!
F = fitted_params(mach)
F.one_hot_encoder

(fitresult = OneHotEncoderResult @566,)

Constructing a linear (unbranching) pipeline with a static (unlearned) target transformation/inverse transformation:

Tree = @load DecisionTreeRegressor pkg=DecisionTree
pipe2 = @pipeline(X -> coerce(X, :age=>Continuous),
                  OneHotEncoder,
                  Tree(max_depth=4),
                  target = y -> log.(y),
                  inverse = z -> exp.(z))

Pipeline275(
    one_hot_encoder = OneHotEncoder(
            features = Symbol[],
            drop_last = false,
            ordered_factor = true,
            ignore = false),
    decision_tree_regressor = DecisionTreeRegressor(
            max_depth = 4,
            min_samples_leaf = 5,
            min_samples_split = 2,
            min_purity_increase = 0.0,
            n_subfeatures = 0,
            post_prune = false,
            merge_purity_threshold = 1.0,
            rng = Random._GLOBAL_RNG()),
    target = WrappedFunction(
            f = Main.ex-workflows.var"#26#27"()),
    inverse = WrappedFunction(
            f = Main.ex-workflows.var"#28#29"())) @972

Creating a homogeneous ensemble of models

Reference: Homogeneous Ensembles

X, y = @load_iris
Tree = @load DecisionTreeClassifier pkg=DecisionTree
tree = Tree()
forest = EnsembleModel(atom=tree, bagging_fraction=0.8, n=300)
mach = machine(forest, X, y)
evaluate!(mach, measure=LogLoss())

PerformanceEvaluation object with these fields:
  measure, measurement, operation, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_pairs
Extract:
┌─────────────────────────────────┬─────────────┬───────────┬───────────────────
│ measure                         │ measurement │ operation │ per_fold         ⋯
├─────────────────────────────────┼─────────────┼───────────┼───────────────────
│ LogLoss(tol = 2.22045e-16) @449 │ 0.625       │ predict   │ [3.66e-15, 3.66e ⋯
└─────────────────────────────────┴─────────────┴───────────┴───────────────────
                                                                1 column omitted

Performance curves

Generate a plot of performance, as a function of some hyperparameter (building on the preceding example)

Single performance curve:

r = range(forest, :n, lower=1, upper=1000, scale=:log10)
curve = learning_curve(mach,
                       range=r,
                       resampling=Holdout(),
                       resolution=50,
                       measure=LogLoss(),
                       verbosity=0)

(parameter_name = "n",
 parameter_scale = :log10,
 parameter_values = [1, 2, 3, 4, 5, 6, 7, 8, 10, 11  …  281, 324, 373, 429, 494, 569, 655, 754, 869, 1000],
 measurements = [4.004850376568572, 7.316553572577199, 4.165577152378119, 2.7699713214474175, 4.2079435052824445, 1.9761433616430988, 1.8511509771176669, 2.8048639450932713, 1.9957133504340576, 2.059054968621477  …  1.2302628173243928, 1.254444233777206, 1.222556599501148, 1.266555369111286, 1.256146669498289, 1.262999185797756, 1.2583000197241907, 1.2457658044223479, 1.2529249911144194, 1.2671784359590974],)

using Plots
plot(curve.parameter_values, curve.measurements, xlab=curve.parameter_name, xscale=curve.parameter_scale)

Multiple curves:

curve = learning_curve(mach,
                       range=r,
                       resampling=Holdout(),
                       measure=LogLoss(),
                       resolution=50,
                       rng_name=:rng,
                       rngs=4,
                       verbosity=0)

(parameter_name = "n",
 parameter_scale = :log10,
 parameter_values = [1, 2, 3, 4, 5, 6, 7, 8, 10, 11  …  281, 324, 373, 429, 494, 569, 655, 754, 869, 1000],
 measurements = [4.004850376568572 8.009700753137142 16.820371581588002 9.611640903764572; 4.004850376568572 7.3473601139354185 2.710975639523342 9.611640903764572; … ; 1.2486785361024142 1.2472072958467921 1.2414807477027914 1.2576209231014552; 1.2510615276661183 1.2519090958212273 1.247134934929548 1.2623870053389732],)

plot(curve.parameter_values, curve.measurements,
xlab=curve.parameter_name, xscale=curve.parameter_scale)