def _handle_value_not_trained_for(self):
"""
Handles the case where a sample has a value for a feature which was
not seen in the training set and therefore is not accounted for in
the tree.
Current strategy is to just return the most common label in the
training data set. It might be better to narrow this down to the
most common among samples that would reach the node at which the
unrecognized value was found.
Returns:
label:
The best guess at the label.
"""
return collection_utils.get_most_common(self.training_set.get_labels())
def _handle_value_not_trained_for(self):
"""
Handles the case where a sample has a value for a feature which was
not seen in the training set and therefore is not accounted for in
the tree.
Current strategy is to just return the most common label in the
training data set. It might be better to narrow this down to the
most common among samples that would reach the node at which the
unrecognized value was found.
Returns:
label:
The best guess at the label.
"""
return collection_utils.get_most_common(
self.training_set.get_labels())
def _build_tree_recursively(dataset):
"""
Private function used to build the decision tree in a recursive fashion.
Args:
dataset: model.DataSet
The data at the current level of the tree. Lower levels of the tree
have filtered subsets of the original data set.
Returns:
current_root: Node
The node which is the root of the level being processed. For
example, on the first/outermost call to this function the root
node will be returned. Subsequent calls will return the various
child nodes.
"""
label_set = set(dataset.get_labels())
if len(label_set) == 1:
# All remaining samples have the same label, no need to split further
return Node(label_set.pop())
if len(dataset.feature_list()) == 0:
# No more features to split on
return Node(get_most_common(dataset.get_labels()))
# We can still split further
split_feature = choose_feature_to_split(dataset)
node = Node(split_feature)
for value in dataset.get_feature_values(split_feature):
subset = dataset.value_filter(
split_feature, value).drop_column(split_feature)
node.add_child(value, _build_tree_recursively(subset))
return node
def _build_tree_recursively(dataset):
"""
Private function used to build the decision tree in a recursive fashion.
Args:
dataset: model.DataSet
The data at the current level of the tree. Lower levels of the tree
have filtered subsets of the original data set.
Returns:
current_root: Node
The node which is the root of the level being processed. For
example, on the first/outermost call to this function the root
node will be returned. Subsequent calls will return the various
child nodes.
"""
label_set = set(dataset.get_labels())
if len(label_set) == 1:
# All remaining samples have the same label, no need to split further
return Node(label_set.pop())
if len(dataset.feature_list()) == 0:
# No more features to split on
return Node(get_most_common(dataset.get_labels()))
# We can still split further
split_feature = choose_feature_to_split(dataset)
node = Node(split_feature)
for value in dataset.get_feature_values(split_feature):
subset = dataset.value_filter(split_feature,
value).drop_column(split_feature)
node.add_child(value, _build_tree_recursively(subset))
return node
def test_get_most_common_empty(self):
collection = []
self.assertIsNone(collection_utils.get_most_common(collection))
def test_get_most_common(self):
collection = ["a", "b", "a", "a", "b"]
self.assertEqual(collection_utils.get_most_common(collection), "a")
def test_get_most_common_empty(self):
collection = []
self.assertIsNone(collection_utils.get_most_common(collection))
def test_get_most_common(self):
collection = ["a", "b", "a", "a", "b"]
self.assertEqual(collection_utils.get_most_common(collection), "a")
