def _get_leaf_nodes(self, node, i_tree, class_label, box):
from copy import deepcopy
from art.metrics.verification_decisions_trees import LeafNode, Box, Interval
leaf_nodes = list()
if 'split_index' in node:
node_left = node['left_child']
node_right = node['right_child']
box_left = deepcopy(box)
box_right = deepcopy(box)
feature = node['split_feature']
box_split_left = Box(intervals={feature: Interval(-np.inf, node['threshold'])})
box_split_right = Box(intervals={feature: Interval(node['threshold'], np.inf)})
if box.intervals:
box_left.intersect_with_box(box_split_left)
box_right.intersect_with_box(box_split_right)
else:
box_left = box_split_left
box_right = box_split_right
leaf_nodes += self._get_leaf_nodes(node_left, i_tree, class_label, box_left)
leaf_nodes += self._get_leaf_nodes(node_right, i_tree, class_label, box_right)
if 'leaf_index' in node:
leaf_nodes.append(LeafNode(tree_id=i_tree, class_label=class_label, node_id=node['leaf_index'], box=box,
value=node['leaf_value']))
return leaf_nodes
def _get_leaf_nodes(self, node_id, i_tree, class_label, box):
from copy import deepcopy
from art.metrics.verification_decisions_trees import LeafNode, Box, Interval
leaf_nodes = list()
if self.get_left_child(node_id) != self.get_right_child(node_id):
node_left = self.get_left_child(node_id)
node_right = self.get_right_child(node_id)
box_left = deepcopy(box)
box_right = deepcopy(box)
feature = self.get_feature_at_node(node_id)
box_split_left = Box(intervals={feature: Interval(-np.inf, self.get_threshold_at_node(node_id))})
box_split_right = Box(intervals={feature: Interval(self.get_threshold_at_node(node_id), np.inf)})
if box.intervals:
box_left.intersect_with_box(box_split_left)
box_right.intersect_with_box(box_split_right)
else:
box_left = box_split_left
box_right = box_split_right
leaf_nodes += self._get_leaf_nodes(node_left, i_tree, class_label, box_left)
leaf_nodes += self._get_leaf_nodes(node_right, i_tree, class_label, box_right)
else:
leaf_nodes.append(LeafNode(tree_id=i_tree, class_label=class_label, node_id=node_id, box=box,
value=self.get_values_at_node(node_id)[0, 0]))
return leaf_nodes
def _get_leaf_nodes(self, node, i_tree, class_label,
box) -> List["LeafNode"]:
from art.metrics.verification_decisions_trees import LeafNode, Box, Interval
leaf_nodes: List[LeafNode] = []
if "children" in node:
if node["children"][0]["nodeid"] == node["yes"] and node[
"children"][1]["nodeid"] == node["no"]:
node_left = node["children"][0]
node_right = node["children"][1]
elif node["children"][1]["nodeid"] == node["yes"] and node[
"children"][0]["nodeid"] == node["no"]:
node_left = node["children"][1]
node_right = node["children"][0]
else:
raise ValueError
box_left = deepcopy(box)
box_right = deepcopy(box)
feature = int(node["split"][1:])
box_split_left = Box(
intervals={
feature: Interval(-np.inf, node["split_condition"])
})
box_split_right = Box(
intervals={feature: Interval(node["split_condition"], np.inf)})
if box.intervals:
box_left.intersect_with_box(box_split_left)
box_right.intersect_with_box(box_split_right)
else:
box_left = box_split_left
box_right = box_split_right
leaf_nodes += self._get_leaf_nodes(node_left, i_tree, class_label,
box_left)
leaf_nodes += self._get_leaf_nodes(node_right, i_tree, class_label,
box_right)
if "leaf" in node:
leaf_nodes.append(
LeafNode(
tree_id=i_tree,
class_label=class_label,
node_id=node["nodeid"],
box=box,
value=node["leaf"],
))
return leaf_nodes
def _get_leaf_nodes(self, node, i_tree, class_label, box):
from copy import deepcopy
from art.metrics.verification_decisions_trees import LeafNode, Box, Interval
leaf_nodes = list()
if 'children' in node:
if node['children'][0]['nodeid'] == node['yes'] and node[
'children'][1]['nodeid'] == node['no']:
node_left = node['children'][0]
node_right = node['children'][1]
elif node['children'][1]['nodeid'] == node['yes'] and node[
'children'][0]['nodeid'] == node['no']:
node_left = node['children'][1]
node_right = node['children'][0]
else:
raise ValueError
box_left = deepcopy(box)
box_right = deepcopy(box)
feature = int(node['split'][1:])
box_split_left = Box(
intervals={
feature: Interval(-np.inf, node['split_condition'])
})
box_split_right = Box(
intervals={feature: Interval(node['split_condition'], np.inf)})
if box.intervals:
box_left.intersect_with_box(box_split_left)
box_right.intersect_with_box(box_split_right)
else:
box_left = box_split_left
box_right = box_split_right
leaf_nodes += self._get_leaf_nodes(node_left, i_tree, class_label,
box_left)
leaf_nodes += self._get_leaf_nodes(node_right, i_tree, class_label,
box_right)
if 'leaf' in node:
leaf_nodes.append(
LeafNode(tree_id=i_tree,
class_label=class_label,
node_id=node['nodeid'],
box=box,
value=node['leaf']))
return leaf_nodes
def get_trees(self):
"""
Get the decision trees.
:return: A list of decision trees.
:rtype: `[Tree]`
"""
import json
from art.metrics.verification_decisions_trees import Box, Tree
booster_dump = self._model.get_booster().get_dump(dump_format="json")
trees = list()
for i_tree, tree_dump in enumerate(booster_dump):
box = Box()
if self._model.n_classes_ == 2:
class_label = -1
else:
class_label = i_tree % self._model.n_classes_
tree_json = json.loads(tree_dump)
trees.append(
Tree(class_id=class_label,
leaf_nodes=self._get_leaf_nodes(tree_json, i_tree,
class_label, box)))
return trees
def get_trees(self) -> list:
"""
Get the decision trees.
:return: A list of decision trees.
"""
from art.metrics.verification_decisions_trees import Box, Tree
booster_dump = self._model.dump_model()["tree_info"]
trees = list()
for i_tree, tree_dump in enumerate(booster_dump):
box = Box()
# pylint: disable=W0212
if self._model._Booster__num_class == 2:
class_label = -1
else:
class_label = i_tree % self._model._Booster__num_class
trees.append(
Tree(
class_id=class_label,
leaf_nodes=self._get_leaf_nodes(
tree_dump["tree_structure"], i_tree, class_label, box),
))
return trees
def get_trees(self):
"""
Get the decision trees.
:return: A list of decision trees.
:rtype: `[Tree]`
"""
from art.metrics.verification_decisions_trees import Box, Tree
trees = list()
for i_tree, decision_tree_model in enumerate(self._model.estimators_):
box = Box()
# if num_classes == 2:
# class_label = -1
# else:
# class_label = i_tree % num_classes
decision_tree_classifier = ScikitlearnDecisionTreeClassifier(model=decision_tree_model)
for i_class in range(self._model.n_classes_):
class_label = i_class
# pylint: disable=W0212
trees.append(Tree(class_id=class_label,
leaf_nodes=decision_tree_classifier._get_leaf_nodes(0, i_tree, class_label, box)))
return trees
def get_trees(self):
"""
Get the decision trees.
:return: A list of decision trees.
:rtype: `[Tree]`
"""
from art.metrics.verification_decisions_trees import Box, Tree
trees = list()
num_trees, num_classes = self._model.estimators_.shape
for i_tree in range(num_trees):
box = Box()
for i_class in range(num_classes):
decision_tree_classifier = ScikitlearnDecisionTreeRegressor(
model=self._model.estimators_[i_tree, i_class])
if num_classes == 2:
class_label = None
else:
class_label = i_class
# pylint: disable=W0212
trees.append(Tree(class_id=class_label,
leaf_nodes=decision_tree_classifier._get_leaf_nodes(0, i_tree, class_label, box)))
return trees
def _get_leaf_nodes(self, node, i_tree, class_label,
box) -> List["LeafNode"]:
from art.metrics.verification_decisions_trees import Box, Interval, LeafNode
leaf_nodes: List[LeafNode] = list()
if "split_index" in node:
node_left = node["left_child"]
node_right = node["right_child"]
box_left = deepcopy(box)
box_right = deepcopy(box)
feature = node["split_feature"]
box_split_left = Box(
intervals={feature: Interval(-np.inf, node["threshold"])})
box_split_right = Box(
intervals={feature: Interval(node["threshold"], np.inf)})
if box.intervals:
box_left.intersect_with_box(box_split_left)
box_right.intersect_with_box(box_split_right)
else:
box_left = box_split_left
box_right = box_split_right
leaf_nodes += self._get_leaf_nodes(node_left, i_tree, class_label,
box_left)
leaf_nodes += self._get_leaf_nodes(node_right, i_tree, class_label,
box_right)
if "leaf_index" in node:
leaf_nodes.append(
LeafNode(
tree_id=i_tree,
class_label=class_label,
node_id=node["leaf_index"],
box=box,
value=node["leaf_value"],
))
return leaf_nodes
