Common MLJ Workflows

Data ingestion

import RDatasets
channing = RDatasets.dataset("boot", "channing")
first(channing, 4)

4 rows × 5 columns

	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, UInt8} │ Multiclass{2} │
│ Entry   │ Int32                           │ Count         │
│ Exit    │ Int32                           │ Count         │
│ Time    │ Int32                           │ Count         │
│ Cens    │ Int32                           │ Count         │
└─────────┴─────────────────────────────────┴───────────────┘
_.nrows = 462

Unpacking data and correcting for wrong scitypes:

y, X =  unpack(channing,
               ==(:Exit),            # y is the :Exit column
               !=(:Time);            # X is the rest, except :Time
               :Exit=>Continuous,
               :Entry=>Continuous,
               :Cens=>Multiclass)
first(X, 4)

4 rows × 3 columns

	Sex	Entry	Cens
	Cat…	Float64	Cat…
1	Male	782.0	1
2	Male	1020.0	1
3	Male	856.0	1
4	Male	915.0	1

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

y[1:4]

4-element Vector{Float64}:
  909.0
 1128.0
  969.0
  957.0

Loading a built-in supervised dataset:

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, :deep_properties, :docstring, :hyperparameter_ranges, :hyperparameter_types, :hyperparameters, :implemented_methods, :is_pure_julia, :is_wrapper, :iteration_parameter, :load_path, :package_license, :package_url, :package_uuid, :prediction_type, :supports_class_weights, :supports_online, :supports_training_losses, :supports_weights, :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,
 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.",
 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],
 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",
 prediction_type = :deterministic,
 supports_class_weights = false,
 supports_online = false,
 supports_training_losses = false,
 supports_weights = false,
 input_scitype = Table{_s46} where _s46<:Union{AbstractVector{_s9} where _s9<:Missing, AbstractVector{_s9} where _s9<:Known},
 target_scitype = AbstractVector{_s349} where _s349<: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) @499

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()])

┌───────────────────────────┬───────────────┬───────────────────────────────┐
│ _.measure                 │ _.measurement │ _.per_fold                    │
├───────────────────────────┼───────────────┼───────────────────────────────┤
│ RootMeanSquaredError @408 │ 8.77          │ [8.53, 8.8, 10.7, 9.43, 5.59] │
│ MeanAbsoluteError @540    │ 6.02          │ [6.52, 5.7, 7.65, 6.09, 4.11] │
└───────────────────────────┴───────────────┴───────────────────────────────┘
_.per_observation = [missing, missing]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

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

import RDatasets
vaso = RDatasets.dataset("robustbase", "vaso"); # a DataFrame
first(vaso, 3)

3 rows × 3 columns

	Volume	Rate	Y
	Float64	Float64	Int64
1	3.7	0.825	1
2	3.5	1.09	1
3	1.25	2.5	1

y, X = unpack(vaso, ==(:Y), c -> true; :Y => Multiclass)

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) @922

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,…} @923 trained 0 times; caches data
  args: 
    1:	Source @902 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @960 ⏎ `AbstractVector{Multiclass{2}}`

Split row indices into training and evaluation rows:

train, test = partition(eachindex(y), 0.7, shuffle=true, rng=1234); # 70:30 split

([27, 28, 30, 31, 32, 18, 21, 9, 26, 14  …  7, 39, 2, 37, 1, 8, 19, 25, 35, 34], [22, 13, 11, 4, 10, 16, 3, 20, 29, 23, 12, 24])

Fit on train and evaluate on test:

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

2.0494478792113626

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 = (Volume=3*rand(3), Rate=3*rand(3))
predict(mach, Xnew)      # a vector of distributions

3-element MLJBase.UnivariateFiniteVector{Multiclass{2}, Int64, UInt32, Float64}:
 UnivariateFinite{Multiclass{2}}(0=>0.0, 1=>1.0)
 UnivariateFinite{Multiclass{2}}(0=>0.9, 1=>0.1)
 UnivariateFinite{Multiclass{2}}(0=>0.273, 1=>0.727)

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

3-element CategoricalArray{Int64,1,UInt32}:
 1
 0
 1

More performance evaluation examples

Evaluating model + data directly:

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

┌──────────────────────────────┬───────────────┬────────────┐
│ _.measure                    │ _.measurement │ _.per_fold │
├──────────────────────────────┼───────────────┼────────────┤
│ LogLoss{Float64} @216        │ 6.52          │ [6.52]     │
│ MarginLoss{ZeroOneLoss} @308 │ 0.417         │ [0.417]    │
└──────────────────────────────┴───────────────┴────────────┘
_.per_observation = [[[0.105, 36.0, ..., 1.3]], [[0.0, 1.0, ..., 1.0]]]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

If a machine is already defined, as above:

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

┌──────────────────────────────┬───────────────┬────────────┐
│ _.measure                    │ _.measurement │ _.per_fold │
├──────────────────────────────┼───────────────┼────────────┤
│ LogLoss{Float64} @216        │ 6.52          │ [6.52]     │
│ MarginLoss{ZeroOneLoss} @308 │ 0.417         │ [0.417]    │
└──────────────────────────────┴───────────────┴────────────┘
_.per_observation = [[[0.105, 36.0, ..., 1.3]], [[0.0, 1.0, ..., 1.0]]]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

Using cross-validation:

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

┌──────────────────────────────┬───────────────┬────────────────────────────────
│ _.measure                    │ _.measurement │ _.per_fold                    ⋯
├──────────────────────────────┼───────────────┼────────────────────────────────
│ LogLoss{Float64} @216        │ 3.27          │ [9.25, 0.598, 4.93, 1.07, 0.5 ⋯
│ MarginLoss{ZeroOneLoss} @308 │ 0.436         │ [0.5, 0.375, 0.375, 0.5, 0.42 ⋯
└──────────────────────────────┴───────────────┴────────────────────────────────
                                                                1 column omitted
_.per_observation = [[[2.22e-16, 0.944, ..., 2.22e-16], [0.847, 0.56, ..., 0.56], [0.799, 0.598, ..., 36.0], [2.01, 2.01, ..., 0.143], [0.847, 2.22e-16, ..., 0.56]], [[0.0, 1.0, ..., 0.0], [1.0, 0.0, ..., 0.0], [1.0, 0.0, ..., 1.0], [1.0, 1.0, ..., 0.0], [1.0, 0.0, ..., 0.0]]]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

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(), ZeroOneLoss()])

┌──────────────────────────────┬───────────────┬───────────────────────┐
│ _.measure                    │ _.measurement │ _.per_fold            │
├──────────────────────────────┼───────────────┼───────────────────────┤
│ LogLoss{Float64} @216        │ 5.88          │ [2.16, 11.0, 4.51]    │
│ MarginLoss{ZeroOneLoss} @308 │ 0.241         │ [0.304, 0.304, 0.115] │
└──────────────────────────────┴───────────────┴───────────────────────┘
_.per_observation = [[[0.154, 0.154, ..., 0.154], [2.22e-16, 36.0, ..., 2.22e-16], [2.22e-16, 2.22e-16, ..., 0.693]], [[0.0, 0.0, ..., 0.0], [0.0, 1.0, ..., 0.0], [0.0, 0.0, ..., 0.0]]]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

Changing a hyperparameter and re-evaluating:

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

┌──────────────────────────────┬───────────────┬────────────────────────────────
│ _.measure                    │ _.measurement │ _.per_fold                    ⋯
├──────────────────────────────┼───────────────┼────────────────────────────────
│ LogLoss{Float64} @216        │ 3.14          │ [9.18, 0.484, 4.86, 0.564, 0. ⋯
│ MarginLoss{ZeroOneLoss} @308 │ 0.361         │ [0.375, 0.25, 0.375, 0.375, 0 ⋯
└──────────────────────────────┴───────────────┴────────────────────────────────
                                                                1 column omitted
_.per_observation = [[[2.22e-16, 1.32, ..., 2.22e-16], [2.22e-16, 0.318, ..., 0.318], [0.575, 2.22e-16, ..., 36.0], [1.5, 1.5, ..., 2.22e-16], [1.22, 2.22e-16, ..., 0.348]], [[0.0, 1.0, ..., 0.0], [0.0, 0.0, ..., 0.0], [0.0, 0.0, ..., 1.0], [1.0, 1.0, ..., 0.0], [1.0, 0.0, ..., 0.0]]]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

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,…} @192 trained 1 time; caches data
  args: 
    1:	Source @954 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @084 ⏎ `AbstractVector{Continuous}`

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

fitted_params(mach)

(coef = [1.0077702564664883, -1.9871151563556402],
 intercept = 0.03912921217390174,)

Get other training-related information:

report(mach)

(deviance = 0.09588865212220386,
 dof_residual = 97.0,
 stderror = [0.010899915987174337, 0.010584502453208538, 0.008191597168575152],
 vcov = [0.0001188081685274587 -5.790229328880606e-6 -5.92796444093823e-5; -5.790229328880606e-6 0.00011203169218197756 -5.26866280949914e-5; -5.92796444093823e-5 -5.26866280949914e-5 6.710226417220846e-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,…} @958 trained 1 time; caches data
  args: 
    1:	Source @325 ⏎ `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 = [4.600156609314096, 4.947780954881536, 4.5162756350682205],
 x2 = [-4.501550627109279, -4.851000786786516, -4.397283748935683],)

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,…} @140.

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),
    atomic_weights = Float64[],
    bagging_fraction = 0.8,
    rng = Random._GLOBAL_RNG(),
    n = 300,
    acceleration = CPU1{Nothing}(nothing),
    out_of_bag_measure = Any[]) @084

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 @266,
            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 = MLJModelInterface.predict,
    range = MLJBase.NumericRange{T, MLJBase.Bounded, Symbol} where T[NumericRange{Float64,…} @802, NumericRange{Int64,…} @620],
    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) @985

Bound the wrapped model to data:

mach = machine(tuned_forest, X, y)

Machine{ProbabilisticTunedModel{Grid,…},…} @662 trained 0 times; caches data
  args: 
    1:	Source @936 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @439 ⏎ `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,…},…} @662 trained 1 time; caches data
  args: 
    1:	Source @936 ⏎ `Table{AbstractVector{Continuous}}`
    2:	Source @439 ⏎ `AbstractVector{Multiclass{3}}`

Inspecting the optimal model:

F = fitted_params(mach)

(best_model = ProbabilisticEnsembleModel{DecisionTreeClassifier} @527,
 best_fitted_params = (fitresult = WrappedEnsemble{Tuple{Node{Float64,…},…},…} @501,),)

F.best_model

ProbabilisticEnsembleModel(
    atom = DecisionTreeClassifier(
            max_depth = -1,
            min_samples_leaf = 1,
            min_samples_split = 2,
            min_purity_increase = 0.0,
            n_subfeatures = 2,
            post_prune = false,
            merge_purity_threshold = 1.0,
            pdf_smoothing = 0.0,
            display_depth = 5),
    atomic_weights = Float64[],
    bagging_fraction = 0.5,
    rng = Random._GLOBAL_RNG(),
    n = 300,
    acceleration = CPU1{Nothing}(nothing),
    out_of_bag_measure = Any[]) @527

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} @416, measure = [BrierScore @135], measurement = [-0.09734266666666652], per_fold = [[0.0, 0.0, -0.1249831111111111, -0.1504026666666662, -0.12969599999999978, -0.17897422222222206]])
 (model = ProbabilisticEnsembleModel{DecisionTreeClassifier} @298, measure = [BrierScore @135], measurement = [-0.11206874074074051], per_fold = [[0.0, 0.0, -0.15049955555555536, -0.15658933333333283, -0.15098933333333295, -0.2143342222222219]])

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=>0.0233, virginica=>0.0, setosa=>0.977)
 UnivariateFinite{Multiclass{3}}(versicolor=>0.0233, virginica=>0.0, setosa=>0.977)
 UnivariateFinite{Multiclass{3}}(versicolor=>0.0233, virginica=>0.0, setosa=>0.977)

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)

Pipeline263(
    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)) @603

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)

┌───────────────────────────┬───────────────┬────────────┐
│ _.measure                 │ _.measurement │ _.per_fold │
├───────────────────────────┼───────────────┼────────────┤
│ RootMeanSquaredError @408 │ 53100.0       │ [53100.0]  │
└───────────────────────────┴───────────────┴────────────┘
_.per_observation = [missing]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

Inspecting the learned parameters in a pipeline:

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

(fitresult = OneHotEncoderResult @103,)

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))

Pipeline274(
    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),
    target = WrappedFunction(
            f = Main.ex-workflows.var"#28#29"()),
    inverse = WrappedFunction(
            f = Main.ex-workflows.var"#30#31"())) @905

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())

┌───────────────────────┬───────────────┬───────────────────────────────────────
│ _.measure             │ _.measurement │ _.per_fold                           ⋯
├───────────────────────┼───────────────┼───────────────────────────────────────
│ LogLoss{Float64} @216 │ 0.426         │ [3.66e-15, 3.66e-15, 0.268, 1.63, 0. ⋯
└───────────────────────┴───────────────┴───────────────────────────────────────
                                                                1 column omitted
_.per_observation = [[[3.66e-15, 3.66e-15, ..., 3.66e-15], [3.66e-15, 3.66e-15, ..., 3.66e-15], [0.0202, 0.00669, ..., 3.66e-15], [3.66e-15, 0.17, ..., 3.66e-15], [3.66e-15, 0.027, ..., 3.66e-15], [0.00669, 0.441, ..., 0.0236]]]
_.fitted_params_per_fold = [ … ]
_.report_per_fold = [ … ]
_.train_test_rows = [ … ]

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, 1.7559728574185285, 8.24744388234224, 4.226889101019484, 7.3283894401232725, 4.290764541094441, 6.555605216786438, 5.0316456080136325, 1.163484509886057, 2.038879929250943  …  1.2422251740406702, 1.236512040182567, 1.2735699053439364, 1.247366859676655, 1.2424995305174167, 1.251580848381322, 1.230963844848692, 1.2446511942872742, 1.238665438807159, 1.2603736522517461],)

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 8.009700753137146 16.820371581588002 8.009700753137146; 4.004850376568572 7.3165535725772 2.710975639523342 8.040507294495367; … ; 1.2518480189324626 1.2441938464592208 1.239609698218395 1.2564846294387948; 1.2522825179547241 1.2532501471645061 1.2412290473284808 1.2540392644156748],)

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