Plotting trees with TreeRecipe.jl

The following example shows how to manually plot the trees learned in DecisionTree.jl:

https://github.com/JuliaAI/TreeRecipe.jl/blob/master/examples/DecisionTree_iris.jl

Currently, the way to integrate a plot recipe in MLJ.jl is not documented, but is sketched in this comment.

So, can we somehow put this together to arrange that a workflow like this generates a plot of a decision tree?

edited again (x2):

using MLJBase
using Plots                 # <---- added in edit
import MLJDecisionTreeInterface
tree = MLJDecisionTreeInterface.DecisionTreeClassifier()
X, y = @load_iris
mach = machine(tree, X, y) |> fit!
plot(mach, 0.8, 0.7; size = (1400,600)))   # <---- added in edit

Note: It used to be that you made RecipesBase.jl your dependency, to avoid a full Plots.jl dependency. But now the recipes live in Plots.jl and you are expected to make Plots.jl a weak dependency. You can see an example of this here.

Start with examples/DecisionTree_iris.jl in TreeRecipe.jl and the MLJBase pull-request comment linked in the issue; compare them with MLJDecisionTreeInterface.DecisionTreeClassifier and the machine/plot example. Done means the shown MLJBase workflow can plot a decision tree, with the Plots integration handled as a weak dependency.