1. class boost:
2.     def __init__(self, tree, n_estimators = 100):
3.         self.models = list()
4.         self.weights = list()
5.         self.errors = list()
6.         self.tree = tree
7.         self.n_estimators = n_estimators
8.         self.sp = 0
9.        
10.     def predict(self, x):
11.         res = np.zeros(x.shape[0])
12.         for model, weight in zip(self.models, self.weights):
13.             res += 1.0 * weight * model.predict(x)
14.         return res
15.    
16.     def _static_predict(self):
17.         return self._sp
18.    
19.     def add_model(self, model, weight, x):
20.         self.models.append(model)
21.         self.weights.append(weight)
22.         self._sp += 1.0 * weight * model.predict(x)
23.        
24.     def _weighted_median(self, a, weights):
25.         sort = np.argsort(a)
26.         a = a[sort]
27.         weights = weights[sort]
28.         weights = weights / np.sum(weights)
29.         weights = np.cumsum(weights)
30.         for i in range(1, a.shape[0] - 1):
31.             if weights[i - 1] <= 1/2 and weights[i + 1] >= 1/2:
32. #                 print(a[i])
33.                 return a[i]
34.  
35.     def fit(self, x, y):
36.         m_0 = self.tree(max_depth = 2)
37.         m_0.fit(x, y)
38.        
39.         self._sp = np.zeros(y.shape[0])
40.         self.add_model(m_0, 1, x)
41.        
42.         for i in range(self.n_estimators):
43. #             print('\r' + str(i))
44.             grad = np.sign(y - self._static_predict())
45.             a = self.tree(max_depth = 3)
46.             a.fit(x, grad)
47.             a_predict = a.predict(x)
48.             old_preds = self._static_predict()
49.            
50.             b = self._weighted_median((y - old_preds) / a_predict, np.abs(a_predict))
51.             self.add_model(a, b, x)
52.            
53.             self.errors.append(np.mean(np.abs(old_preds - y)))
54.    
55.     def get_errors(self):
56.         return self.errors