def build_tree(self, x_train: pd.DataFrame, y_train: np.array,
tree: TreeNode) -> None:
"""
Build decision tree with ID3 feature choosing algorithm.
:param x_train: Pandas data frame with the data (features) to train the model on. The array length is N*M where
N is the number of samples in the dataset, and M is the number of features.
:param y_train: Numpy array with the labels of the data. Length is N.
:param tree: A root node.
:return: None.
"""
# Create new leaf with the class.
if x_train.shape[0] == 1 or y_train.shape[
0] <= self.pruning_M or self.are_all_labels_equal(y_train):
new_node = TreeNode(self.find_major_class(y_train))
tree.add_child(new_node)
return
# Find best feature to continue with, and best threhold to split the data with it.
f_name, best_threshold = self.get_feature_with_max_info_gain(
x_train, y_train)
# Add another layer to the tree representing the current f, add thresh for prediction later.
new_node = TreeNode((f_name, str(best_threshold)))
tree.add_child(new_node)
samples_great, labels_great, samples_small, labels_small = self.split_by_threshold(
data=x_train, threshold=best_threshold, feature_name=f_name)
if samples_great.shape[0] == 1:
# If there is only one sample, no need to do recursion call.
node = TreeNode(labels_great[0])
new_node.add_child(node)
else:
self.build_tree(samples_great, labels_great, new_node)
if samples_small.shape[0] == 1:
# If there is only one sample, no need to do recursion call.
node = TreeNode(labels_small[0])
new_node.add_child(node)
else:
self.build_tree(samples_small, labels_small, new_node)
return
def _build_parent_node_list(directory, level, children):
node_list = list()
index = 0
children.sort(key=lambda child_node: child_node.get_path())
while index < len(children):
if index % 2 != 0:
node = TreeNode(directory + '_' + level + '_' + (index + 1) % 2)
node.add_child(children[index])
node.add_child(children[index - 1])
if index == len(children) - 2:
index += 1
node.add_child(children[index])
node.calc_signature()
node_list.append(node)
index += 1
return node_list
def _create_dir_node(dir_path, node_list):
node = TreeNode(dir_path)
for child_node in node_list:
node.add_child(child_node)
node.calc_signature()
return node
