import numpy as np
import pandas as pd
class MyTreeClf:
def __init__(self, max_depth=5, min_samples_split=2, max_leafs=20):
self.max_depth = max_depth
self.min_samples_split = min_samples_split
self.max_leafs = max_leafs
self.tree = None
self.leafs_cnt = 0
def node_entropy(self, probs):
return -np.sum([p * np.log2(p) for p in probs if p > 0])
def node_ig(self, x_col, y, split_value):
left_mask = x_col <= split_value
right_mask = x_col > split_value
if len(x_col[left_mask]) == 0 or len(x_col[right_mask]) == 0:
return 0
left_counts = np.bincount(y[left_mask])
right_counts = np.bincount(y[right_mask])
left_probs = left_counts / len(y[left_mask]) if len(y[left_mask]) > 0 else np.zeros_like(left_counts)
right_probs = right_counts / len(y[right_mask]) if len(y[right_mask]) > 0 else np.zeros_like(right_counts)
entropy_after = (len(y[left_mask]) / len(y) * self.node_entropy(left_probs) +
len(y[right_mask]) / len(y) * self.node_entropy(right_probs))
entropy_before = self.node_entropy(np.bincount(y) / len(y))
return entropy_before - entropy_after
def get_best_split(self, X: pd.DataFrame, y: pd.Series):
best_col, best_split_value, best_ig = None, None, -np.inf
for col in X.columns:
sorted_unique_values = np.sort(X[col].unique())
for i in range(1, len(sorted_unique_values)):
split_value = (sorted_unique_values[i - 1] + sorted_unique_values[i]) / 2
ig = self.node_ig(X[col], y, split_value)
if ig > best_ig:
best_ig = ig
best_col = col
best_split_value = split_value
return best_col, best_split_value
def fit(self, X: pd.DataFrame, y: pd.Series, depth=1, node=None):
if self.max_leafs < 2:
self.leafs_cnt = 2
if node is None:
node = {}
self.tree = node
best_col, best_split_value = self.get_best_split(X, y)
node['type'] = None
node['feature'] = best_col
node['threshold'] = best_split_value
tmp_leafs = 2
if len(y.unique()) == 1:
self.leafs_cnt += 1
tmp_leafs -= 1
node['type'] = 'leaf'
node['class_counts'] = dict(zip(*np.unique(y, return_counts=True)))
return
if len(y) == 1:
self.leafs_cnt += 1
tmp_leafs -= 1
node['type'] = 'leaf'
node['class_counts'] = dict(zip(*np.unique(y, return_counts=True)))
return
if depth > self.max_depth or len(y) < self.min_samples_split or (tmp_leafs + self.leafs_cnt) > self.max_leafs:
self.leafs_cnt += 1
tmp_leafs -= 1
node['type'] = 'leaf'
node['class_counts'] = dict(zip(*np.unique(y, return_counts=True)))
return
if best_col is None:
self.leafs_cnt += 1
tmp_leafs -= 1
node['type'] = 'leaf'
node['class_counts'] = dict(zip(*np.unique(y, return_counts=True)))
return
node['type'] = 'node'
node['feature'] = best_col
node['threshold'] = best_split_value
tmp_leafs += 2
left_mask = X[best_col] <= best_split_value
right_mask = X[best_col] > best_split_value
node['left'] = {}
node['right'] = {}
self.fit(X[left_mask], y[left_mask], depth + 1, node['left'])
self.fit(X[right_mask], y[right_mask], depth + 1, node['right'])
def bypass_tree(self, node, sample):
while node['type'] == 'node':
feature_value = sample[node['feature']]
if feature_value <= node['threshold']:
node = node['left']
else:
node = node['right']
if node['type'] == 'leaf':
return node
def predict(self, X: pd.DataFrame):
classification = []
proba = self.predict_proba(X)
for p in proba:
if p > 0.5:
classification.append(1)
else:
classification.append(0)
return classification
def predict_proba(self, X: pd.DataFrame):
proba = []
for _, sample in X.iterrows():
node = self.bypass_tree(self.tree, sample)
if node:
overall_cnt = sum(node['class_counts'].values())
class1_cnt = node['class_counts'].get(1, 0)
class1_proba = (class1_cnt / overall_cnt) if overall_cnt > 0 else 0.0
proba.append(class1_proba)
return np.array(proba)
df = pd.read_csv('c:\\Users\\Nijat\\Downloads\\banknote+authentication.zip', header=None)
df.columns = ['variance', 'skewness', 'curtosis', 'entropy', 'target']
X, y = df.iloc[:, :4], df['target']
model1 = MyTreeClf(max_depth=3, min_samples_split=2, max_leafs=5)
model1.fit(X, y)
result1 = model1.predict_proba(X)
print(result1)
from sklearn.tree import DecisionTreeClassifier
model2 = DecisionTreeClassifier(max_depth=3, min_samples_split=2, max_leaf_nodes=5)
model2.fit(X, y)
result2 = model2.predict_proba(X)
print(result2[:, 1])
Здравствуйте. Пытаюсь сделать методы '''predict()''' и '''predict_proba()''' и когда сверяю результаты моего '''predict_proba()''' и в реализации от sklearn, то значения не совпадают. Я предполагаю что проблема либо в неправильном подсчете вероятностей, либо же в неправильных сплитах данных.
Проверка
Входные данные: два набора параметров для дерева решений
Выходные данные: возвращенные предсказания (их сумма) вероятностей и меток
Sample Input:
{"max_depth": 3, "min_samples_split": 2, "max_leafs": 1}
Sample Output:
11.2443438914 (у меня был ответ 10.0)
К сожалению, того самого датасета у меня нет, но я использую для проверки sklearn и датасет "Banknote authentication"
Вот ссылка:
https://archive.ics.uci.edu/static/public/267/bank...
Подробнее про задачу:
https://stepik.org/lesson/333977/step/7?unit=317388
Простой