def decision_tree_algorithm(df, counter=0, min_samples=2, max_depth=5, random_subspace=None):
# data preparations
if counter == 0:
global COLUMN_HEADERS, FEATURE_TYPES
COLUMN_HEADERS = df.columns
FEATURE_TYPES = determine_type_of_feature(df)
data = df.values
else:
data = df
# base cases
if (check_purity(data)) or (len(data) < min_samples) or (counter == max_depth):
classification = classify_data(data)
return classification
# recursive part
else:
counter += 1
# helper functions
potential_splits = get_potential_splits(data, random_subspace)
split_column, split_value = determine_best_split(data, potential_splits)
data_below, data_above = split_data(data, split_column, split_value)
# check for empty data
if len(data_below) == 0 or len(data_above) == 0:
classification = classify_data(data)
return classification
# determine question
feature_name = COLUMN_HEADERS[split_column]
type_of_feature = FEATURE_TYPES[split_column]
if type_of_feature == "continuous":
question = "{} <= {}".format(feature_name, split_value)
# feature is categorical
else:
question = "{} = {}".format(feature_name, split_value)
# instantiate sub-tree
sub_tree = {question: []}
# find answers (recursion)
yes_answer = decision_tree_algorithm(data_below, counter, min_samples, max_depth, random_subspace)
no_answer = decision_tree_algorithm(data_above, counter, min_samples, max_depth, random_subspace)
# If the answers are the same, then there is no point in asking the qestion.
# This could happen when the data is classified even though it is not pure
# yet (min_samples or max_depth base case).
if yes_answer == no_answer:
sub_tree = yes_answer
else:
sub_tree[question].append(yes_answer)
sub_tree[question].append(no_answer)
return sub_tree
def decision_tree_algorithm(df, counter=0, min_samples=2, max_depth=5, random_subspace=None):
# data preparations
if counter == 0:
global COLUMN_HEADERS, FEATURE_TYPES
COLUMN_HEADERS = df.columns
FEATURE_TYPES = determine_type_of_feature(df)
data = df.values
else:
data = df
if (check_purity(data)) or (len(data) < min_samples) or (counter == max_depth):
classification = classify_data(data)
return classification
else:
counter += 1
potential_splits = get_potential_splits(data, random_subspace)
split_column, split_value = determine_best_split(data, potential_splits)
data_below, data_above = split_data(data, split_column, split_value)
if len(data_below) == 0 or len(data_above) == 0:
classification = classify_data(data)
return classification
feature_name = COLUMN_HEADERS[split_column]
type_of_feature = FEATURE_TYPES[split_column]
if type_of_feature == "continuous":
question = "{} <= {}".format(feature_name, split_value)
else:
question = "{} = {}".format(feature_name, split_value)
sub_tree = {question: []}
yes_answer = decision_tree_algorithm(data_below, counter, min_samples, max_depth, random_subspace)
no_answer = decision_tree_algorithm(data_above, counter, min_samples, max_depth, random_subspace)
if yes_answer == no_answer:
sub_tree = yes_answer
else:
sub_tree[question].append(yes_answer)
sub_tree[question].append(no_answer)
return sub_tree