class TestTreeStructureDataUpdate(TestCase):
def setUp(self):
X = format_covariate_matrix(
pd.DataFrame({
"a": [1, 2, 3],
"b": [1, 2, 3]
}))
self.data = Data(X, np.array([1, 2, 3]).astype(float))
self.a = split_node(LeafNode(Split(
self.data)), (SplitCondition(0, 1, le), SplitCondition(0, 1, gt)))
self.b = self.a.left_child
self.x = self.a.right_child
self.tree = Tree([self.a, self.b, self.x])
self.c = split_node(
self.a._right_child,
(SplitCondition(1, 2, le), SplitCondition(1, 2, gt)))
mutate(self.tree, TreeMutation("grow", self.x, self.c))
self.d = self.c.left_child
self.e = self.c.right_child
def test_update_pushed_through_split(self):
updated_y = np.array([5, 6, 7])
self.tree.update_y(updated_y)
# Left child keeps LTE condition
self.assertListEqual([5, 6, 7], list(self.a.data.y))
self.assertListEqual([5], list(self.b.data.y.compressed()))
self.assertListEqual([6, 7], list(self.c.data.y.compressed()))
self.assertListEqual([6], list(self.d.data.y.compressed()))
self.assertListEqual([7], list(self.e.data.y.compressed()))
def test_same_prediction(self):
from sklearn.ensemble import GradientBoostingRegressor
params = {
'n_estimators': 1,
'max_depth': 2,
'min_samples_split': 2,
'learning_rate': 0.8,
'loss': 'ls'
}
sklearn_model = GradientBoostingRegressor(**params)
sklearn_model.fit(self.data.X.values, self.data.y.values)
sklearn_tree = sklearn_model.estimators_[0][0].tree_
bartpy_tree = Tree([LeafNode(Split(self.data))])
map_sklearn_tree_into_bartpy(bartpy_tree, sklearn_tree)
sklearn_predictions = sklearn_tree.predict(
self.data.X.values.astype(np.float32))
sklearn_predictions = [
round(x, 2) for x in sklearn_predictions.reshape(-1)
]
bartpy_tree.cache_up_to_date = False
bartpy_tree_predictions = bartpy_tree.predict(self.data.X.values)
bartpy_tree_predictions = [
round(x, 2) for x in bartpy_tree_predictions
]
self.assertListEqual(sklearn_predictions, bartpy_tree_predictions)
def initialize_trees(self) -> List[Tree]:
tree_data = deepcopy(self.data)
tree_data._y = tree_data.y / self.n_trees
trees = [
Tree([LeafNode(Split(self.data))]) for _ in range(self.n_trees)
]
return trees
def initialize_trees(self) -> List[Tree]:
tree_data = copy(self.data)
tree_data.update_y(tree_data.y / self.n_trees)
trees = [
Tree([LeafNode(Split(tree_data))]) for _ in range(self.n_trees)
]
return trees
def setUp(self):
self.data = Data(pd.DataFrame({"a": [1, 2]}).values, np.array([1, 1]))
self.d = LeafNode(Split(self.data))
self.e = LeafNode(Split(self.data))
self.c = DecisionNode(Split(self.data), self.d, self.e)
self.b = LeafNode(Split(self.data))
self.a = DecisionNode(Split(self.data), self.b, self.c)
self.tree = Tree([self.a, self.b, self.c, self.d, self.e])
def initialize_trees(self) -> List[Tree]:
trees = [
Tree([LeafNode(Split(deepcopy(self.data)))])
for _ in range(self.n_trees)
]
for tree in trees:
tree.update_y(tree.update_y(self.data.y.values / self.n_trees))
return trees
def test_head_prune(self):
b, c = LeafNode(Split(self.data)), LeafNode(Split(self.data))
a = DecisionNode(Split(self.data), b, c)
tree = Tree([a, b, c])
updated_a = LeafNode(Split(self.data))
prune_mutation = PruneMutation(a, updated_a)
mutate(tree, prune_mutation)
self.assertIn(updated_a, tree.leaf_nodes)
self.assertNotIn(self.a, tree.nodes)
def setUp(self):
self.data = Data(format_covariate_matrix(pd.DataFrame({"a": [1]})),
np.array([1]).astype(float))
self.d = LeafNode(Split(self.data), None)
self.e = LeafNode(Split(self.data), None)
self.c = DecisionNode(Split(self.data), self.d, self.e)
self.b = LeafNode(Split(self.data))
self.a = DecisionNode(Split(self.data), self.b, self.c)
self.tree = Tree([self.a, self.b, self.c, self.d, self.e])
def setUp(self):
self.data = make_bartpy_data(
pd.DataFrame({"a": np.random.normal(size=1000)}),
np.array(np.random.normal(size=1000)))
self.d = LeafNode(Split(self.data))
self.e = LeafNode(Split(self.data))
self.c = DecisionNode(Split(self.data), self.d, self.e)
self.b = LeafNode(Split(self.data))
self.a = DecisionNode(Split(self.data), self.b, self.c)
self.tree = Tree([self.a, self.b, self.c, self.d, self.e])
def setUp(self):
self.data = make_bartpy_data(pd.DataFrame({"a": [1, 2]}),
np.array([1, 2]),
normalize=False)
self.d = LeafNode(Split(self.data))
self.e = LeafNode(Split(self.data))
self.c = DecisionNode(Split(self.data), self.d, self.e)
self.b = LeafNode(Split(self.data))
self.a = DecisionNode(Split(self.data), self.b, self.c)
self.tree = Tree([self.a, self.b, self.c, self.d, self.e])
def setUp(self):
self.data = Data(
pd.DataFrame({
"a": [1, 2, 3],
"b": [1, 2, 3]
}).values, np.array([1, 2, 3]))
self.a = split_node(LeafNode(Split(
self.data)), (SplitCondition(0, 1, le), SplitCondition(0, 1, gt)))
self.b = self.a.left_child
self.x = self.a.right_child
self.tree = Tree([self.a, self.b, self.x])
self.c = split_node(
self.a._right_child,
(SplitCondition(1, 2, le), SplitCondition(1, 2, gt)))
mutate(self.tree, TreeMutation("grow", self.x, self.c))
self.d = self.c.left_child
self.e = self.c.right_child
def setUp(self):
X = format_covariate_matrix(
pd.DataFrame({
"a": [1, 2, 3],
"b": [1, 2, 3]
}))
self.data = Data(X, np.array([1, 2, 3]).astype(float))
self.a = split_node(LeafNode(Split(
self.data)), (SplitCondition(0, 1, le), SplitCondition(0, 1, gt)))
self.b = self.a.left_child
self.x = self.a.right_child
self.tree = Tree([self.a, self.b, self.x])
self.c = split_node(
self.a._right_child,
(SplitCondition(1, 2, le), SplitCondition(1, 2, gt)))
mutate(self.tree, TreeMutation("grow", self.x, self.c))
self.d = self.c.left_child
self.e = self.c.right_child
