Learners

Similar to SKPreprocessor, most of the Learners in AMLP for its initial release are based on Scikitlearn libraries.

SKLearner Structure

    SKLearner(Dict(
       :name => "sklearner",
       :output => :class,
       :learner => "LinearSVC",
       :impl_args => Dict()
      )
    )

Helper Function:
    SKLearner(learner::String,args::Dict=Dict())

SKLearner maintains a dictionary of learners which can be listed by invoking the function: sklearners() The :impl_args is a dictionary of paramters to be passed as arguments to the Scikitlearn learner.

Let's try loading some learners with some arguments based on Scikitlearn documentation:

julia> using AutoMLPipeline

julia> iris = getiris();

julia> X = iris[:,1:4];

julia> Y = iris[:,end] |> Vector;

julia> rf = SKLearner("RandomForestClassifier",Dict(:n_estimators=>30,:random_state=>0));

julia> crossvalidate(rf,X,Y,"accuracy_score",3)
fold: 1, 0.94
fold: 2, 0.92
fold: 3, 0.96
errors: 0
(mean = 0.94, std = 0.019999999999999962, folds = 3, errors = 0)

julia> ada = SKLearner("AdaBoostClassifier",Dict(:n_estimators=>20,:random_state=>0));

julia> crossvalidate(ada,X,Y,"accuracy_score",3)
fold: 1, 0.92
fold: 2, 0.94
fold: 3, 0.98
errors: 0
(mean = 0.9466666666666667, std = 0.030550504633038912, folds = 3, errors = 0)

julia> svc = SKLearner("SVC",Dict(:kernel=>"rbf",:random_state=>0,:gamma=>"auto"));

julia> crossvalidate(svc,X,Y,"accuracy_score",3)
fold: 1, 0.94
fold: 2, 0.96
fold: 3, 0.94
errors: 0
(mean = 0.9466666666666667, std = 0.011547005383792525, folds = 3, errors = 0)