Refactor ComponentSelector to match sklearn's DecisionTreeClassifier

[tsalo]

Summary

If we use DecisionTreeClassifier as the general template for ComponentSelector.

• __init__() takes hyperparameters, but not data. In our case, that would be the tree, but not the comptable.
• fit() takes training data and fits the model. In our case, comptable is training data.
• predict() takes test data and predicts labels. In our case, comptable is the test data as well.

If we organize the tree's attributes similarly to DecisionTreeClassifier, we might be able to directly use sklearn's plot_tree.
It would also be really cool if we could figure out how to plot the decision surfaces based on features (see this example).