Дерево решений является одним из наиболее широко используемых алгоритмов машинного обучения, поскольку они просты для понимания и интерпретации, просты в использовании, универсальны и эффективны.
В этой статье мы рассмотрим различные способы визуализации дерева решений в Python.
Сначала мы создадим простое дерево решений, используя набор данных IRIS.
from sklearn.datasets import load_iris from sklearn import tree import matplotlib % matplotlib inline iris = load_iris() clf = tree.DecisionTreeClassifier() clf = clf.fit(iris.data, iris.target)
sklearn.tree предоставляет функцию дерева графиков для визуализации дерева в блокноте Jupyter.
data =tree.plot_tree(clf.fit(iris.data, iris.target),filled=True)
Также вы можете сохранить визуализацию в формате graphviz как:
dotfile = open(“DT_metrics.dot”, ‘w’) tree.export_graphviz(clf, out_file=dotfile, feature_names=iris.feature_names,class_names=iris.target_names) dotfile.close()
Чтобы визуализировать файл .dot, скопируйте его содержимое на http://webgraphviz.com/ и сгенерируйте дерево.
sklearn.treeпредоставляет функцию export_text для получения английских правил дерева.
from sklearn.tree.export import export_text r = export_text(clf, feature_names=iris[‘feature_names’]) print(r) |--- petal length (cm) <= 2.45 | |--- class: 0 |--- petal length (cm) > 2.45 | |--- petal width (cm) <= 1.75 | | |--- petal length (cm) <= 4.95 | | | |--- petal width (cm) <= 1.65 | | | | |--- class: 1 | | | |--- petal width (cm) > 1.65 | | | | |--- class: 2 | | |--- petal length (cm) > 4.95 | | | |--- petal width (cm) <= 1.55 | | | | |--- class: 2 | | | |--- petal width (cm) > 1.55 | | | | |--- petal length (cm) <= 5.45 | | | | | |--- class: 1 | | | | |--- petal length (cm) > 5.45 | | | | | |--- class: 2 | |--- petal width (cm) > 1.75 | | |--- petal length (cm) <= 4.85 | | | |--- sepal width (cm) <= 3.10 | | | | |--- class: 2 | | | |--- sepal width (cm) > 3.10 | | | | |--- class: 1 | | |--- petal length (cm) > 4.85 | | | |--- class: 2
Вы также можете написать пользовательскую функцию, как показано ниже, чтобы получить правила в текстовом формате.
def tree_rules(dt_model, target_names, feature_names):
"""
Return the tree if-then-else rules
Parameters:
-----------
dt_model : DecisionTreeClassifier object
target_names : Target class names (Y)
feature_names : Features / Predictors (X)
Returns:
--------
tree_rules: Decision tree rules
Example:
--------
>>> from sklearn.datasets import load_iris
>>> from sklearn import tree
>>> iris = load_iris()
>>> dt_model = tree.DecisionTreeClassifier().fit(iris.data, iris.target)
>>> print(tree_rules(dt_model, iris.target_names,iris.feature_names))
"""
if (dt_model is None) or (feature_names is None) or (target_names is None):
raise ValueError("Please pass DecisionTreeClassifier object, list of features and target variable classes")
tree_ = dt_model.tree_
tree_rules = ""
left = dt_model.tree_.children_left
right = dt_model.tree_.children_right
threshold = dt_model.tree_.threshold
features = [feature_names[i] for i in dt_model.tree_.feature]
value = dt_model.tree_.value
tree_rules = tree_rules + "\n"+ "Structure of the tree using Predictors:" + format(", ".join(feature_names))
def add_node(left, right, threshold, features, node, depth=0):
indent = " " * depth
tree_rules = ""
if (threshold[node] != -2):
tree_rules = tree_rules+ "\n"+ indent + "if ( " + features[node] + " <= " + str(threshold[node]) + " ) {"
if left[node] != -1:
tree_rules = tree_rules + add_node (left, right, threshold, features, left[node], depth+1)
tree_rules = tree_rules+"\n" + indent + "} else {"
if right[node] != -1:
tree_rules = tree_rules + add_node (left, right, threshold, features, right[node], depth+1)
tree_rules = tree_rules+"\n"+ indent + "}"
else:
tree_rules = "\n" + tree_rules+ indent + "Prediction: " + target_names[np.argmax(tree_.value[node][0])]
return tree_rules
tree_rules = tree_rules + add_node(left, right, threshold, features, 0)
return tree_rules
Чтобы получить дерево решений в виде JSON, которое можно использовать для рендеринга в веб-приложении:
def tree_to_json(dt_model, target_names, feature_names=None):
"""
Returns the Decision Tree output rules in JSON format
Parameters:
-----------
dt_model : DecisionTreeClassifier object
target_names : Target class names (Y)
feature_names : Features / Predictors (X)
Returns:
--------
tree_json: Decision tree output as JSON
Example
--------
>>> from sklearn.datasets import load_iris
>>> from sklearn import tree
>>> iris = load_iris()
>>> dt_model = tree.DecisionTreeClassifier().fit(iris.data, iris.target)
>>> print(tree_to_json(dt_model,iris.target_names,iris.feature_names))
"""
if (dt_model is None) or (feature_names is None) or (target_names is None):
raise ValueError("Please pass DecisionTreeClassifier object, list of features and target variable classes")
tree_ = dt_model.tree_
tree_json = ""
def node_to_str(tree, node_id, criterion):
if not isinstance(criterion, six.string_types):
criterion = "impurity"
value = tree.value[node_id]
if tree.n_outputs == 1:
value = value[0, :]
jsonValue = ', '.join([str(x) for x in value])
if tree.children_left[node_id] == sklearn.tree._tree.TREE_LEAF:
is_leaf = 'TRUE'
return '"id": "%s", "isLeaf": "%s", "criterion": "%s", "impurity": "%s", "samples": "%s", "prediction": "%s", "samplesDistribution": [%s]' \
% (node_id,
is_leaf,
criterion,
tree.impurity[node_id],
tree.n_node_samples[node_id],
target_names[np.argmax(tree_.value[node_id][0])],
jsonValue)
else:
is_leaf = 'FALSE'
if feature_names is not None:
feature = feature_names[tree.feature[node_id]]
else:
feature = tree.feature[node_id]
if "=" in feature:
ruleType = "="
ruleValue = "false"
else:
ruleType = "<="
ruleValue = "%.3f" % tree.threshold[node_id]
return '"id": "%s", "isLeaf": "%s", "rule": "%s %s %s", "impurity": "%s", "samples": "%s"' \
% (node_id,
is_leaf,
feature,
ruleType,
ruleValue,
tree.impurity[node_id],
tree.n_node_samples[node_id])
def add_node(tree, node_id, criterion, parent=None, depth=0):
tabs = " " * depth
tree_json = ""
left_child = tree.children_left[node_id]
right_child = tree.children_right[node_id]
tree_json = tree_json + "\n" + tabs + "{\n" + tabs + " " + node_to_str(tree, node_id, criterion)
if left_child != sklearn.tree._tree.TREE_LEAF:
tree_json = tree_json + ",\n" + tabs + ' "left": ' + \
add_node(tree, left_child, criterion=criterion, parent=node_id, depth=depth + 1) + \
",\n" + tabs + ' "right": ' + \
add_node(tree, right_child, criterion=criterion, parent=node_id, depth=depth + 1)
tree_json = tree_json + tabs + "\n" + tabs + "}"
return tree_json
if isinstance(dt_model, sklearn.tree.tree.Tree):
tree_json = tree_json + add_node(dt_model, 0, criterion="gini")
else:
tree_json = tree_json + add_node(dt_model.tree_, 0, criterion=dt_model.criterion)
return tree_json
Чтобы получить дерево решений в виде упорядоченного словаря:
def tree_to_dict(dt_model, target_names, feature_names=None):
"""
Returns the Decision Tree output rules as dictionary
Parameters:
-----------
dt_model : DecisionTreeClassifier object
target_names : Target class names (Y)
feature_names : Features / Predictors (X)
Returns:
--------
tree_dict: Decision tree output as dictionary
Example
--------
>>> from sklearn.datasets import load_iris
>>> from sklearn import tree
>>> iris = load_iris()
>>> dt_model = tree.DecisionTreeClassifier().fit(iris.data, iris.target)
>>> print(tree_to_dict(dt_model,iris.target_names,iris.feature_names))
"""
tree_ = dt_model.tree_
tree_dict = OrderedDict()
def node_to_str(tree, node_id, criterion):
if not isinstance(criterion, six.string_types):
criterion = "impurity"
value = tree.value[node_id]
if tree.n_outputs == 1:
value = value[0, :]
json_value = ', '.join([str(x) for x in value])
if tree.children_left[node_id] == sklearn.tree._tree.TREE_LEAF:
is_leaf = 'TRUE'
return {"id": int(node_id), "isLeaf": is_leaf, "criterion": criterion, "impurity": float(tree.impurity[node_id]), "samples": float(tree.n_node_samples[node_id]), "prediction": str(target_names[np.argmax(tree_.value[node_id][0])]), "samplesDistribution": json_value}
else:
is_leaf = 'FALSE'
if feature_names is not None:
feature = str(feature_names[tree.feature[node_id]])
else:
feature = tree.feature[node_id]
if "=" in feature:
rule_type = "="
rule_value = "false"
else:
rule_type = "<="
rule_value = "%.3f" % tree.threshold[node_id]
return {"id": int(node_id), "isLeaf": is_leaf, "rule": feature + rule_type + rule_value, "impurity": float(tree.impurity[node_id]),
"samples": float(tree.n_node_samples[node_id])}
def add_node(tree, node_id, criterion, parent=None, depth=0):
tree_dict = OrderedDict()
left_child = tree.children_left[node_id]
right_child = tree.children_right[node_id]
tree_dict.update(node_to_str(tree, node_id, criterion))
if left_child != sklearn.tree._tree.TREE_LEAF:
tree_dict.update({"left": add_node(tree, left_child, criterion=criterion, parent=node_id, depth=depth + 1)})
tree_dict.update({"right": add_node(tree, right_child, criterion=criterion, parent=node_id, depth=depth + 1)})
return tree_dict
if isinstance(dt_model, sklearn.tree.tree.Tree):
tree_dict.update(add_node(dt_model, 0, criterion="gini"))
else:
tree_dict.update(add_node(dt_model.tree_, 0, criterion=dt_model.criterion))
return tree_dict
