Tree

class bartpy.tree.Tree(nodes: typing.List[bartpy.node.TreeNode])

An encapsulation of the structure of a single decision tree Contains no logic, but keeps track of 4 different kinds of nodes within the tree:

• leaf nodes
• decision nodes
• splittable leaf nodes
• prunable decision nodes

Parameters:nodes (List[Node]) – All nodes contained in the tree, i.e. decision and leaf nodes

add_node(node: bartpy.node.TreeNode) → None

Add a node to the tree Note that this is non-recursive, only adds the node and not any children

decision_nodes

List of decision nodes in the tree. Decision nodes are internal split nodes, i.e. not leaf nodes

leaf_nodes

List of all of the leaf nodes in the tree

nodes

List of all nodes contained in the tree

out_of_sample_predict(X) → numpy.ndarray

Prediction for a covariate matrix not used for training

Note that this is quite slow

Parameters:X (pd.DataFrame) – Covariates to predict for Returns: Return type:np.ndarray

predict() → numpy.ndarray

Generate a set of predictions with the same dimensionality as the target array Note that the prediction is from one tree, so represents only (1 / number_of_trees) of the target

prunable_decision_nodes

List of decision nodes in the tree that are suitable for pruning In particular, decision nodes that have two leaf node children

remove_node(node: bartpy.node.TreeNode) → None

Remove a single node from the tree Note that this is non-recursive, only drops the node and not any children

splittable_leaf_nodes

List of all leaf nodes in the tree which can be split in a non-degenerate way i.e. not all rows of the covariate matrix are duplicates

update_y(y: numpy.ndarray) → None

Update the cached value of the target array in all nodes Used to pass in the residuals from the sum of all of the other trees

bartpy.tree.mutate(tree: bartpy.tree.Tree, mutation: bartpy.mutation.TreeMutation) → None

Apply a change to the structure of the tree Modifies not only the tree, but also the links between the TreeNodes

Parameters:

• tree (Tree) – The tree to mutate
• mutation (TreeMutation) – The mutation to apply to the tree

class bartpy.mutation.TreeMutation(kind: str, existing_node: bartpy.node.TreeNode, updated_node: bartpy.node.TreeNode)

An encapsulation of a change to be made to the tree. Constructed of three components

• the node to be changed
• what it should be changed to
• a string name of the kind of change (normally grow or prune)