Coverage for mlinsights/mlmodel/piecewise_tree_regression.py: 99%

1# -*- coding: utf-8 -*- 

2""" 

3@file 

4@brief Implements a kind of piecewise linear regression by modifying 

5the criterion used by the algorithm which builds a decision tree. 

6""" 

7import numpy 

8from sklearn.tree import DecisionTreeRegressor 

9 

10 

11class PiecewiseTreeRegressor(DecisionTreeRegressor): 

12 """ 

13 Implements a kind of piecewise linear regression by modifying 

14 the criterion used by the algorithm which builds a decision tree. 

15 See :epkg:`sklearn:tree:DecisionTreeRegressor` to get the meaning 

16 of the parameters except criterion: 

17 

18 * ``mselin``: optimizes for a piecewise linear regression 

19 * ``simple``: optimizes for a stepwise regression (equivalent to *mse*) 

20 """ 

21 

22 def __init__(self, criterion='mselin', splitter='best', max_depth=None, 

23 min_samples_split=2, min_samples_leaf=1, 

24 min_weight_fraction_leaf=0.0, max_features=None, 

25 random_state=None, max_leaf_nodes=None, 

26 min_impurity_decrease=0.0): 

27 DecisionTreeRegressor.__init__( 

28 self, criterion=criterion, 

29 splitter=splitter, max_depth=max_depth, 

30 min_samples_split=min_samples_split, 

31 min_samples_leaf=min_samples_leaf, 

32 min_weight_fraction_leaf=min_weight_fraction_leaf, 

33 max_features=max_features, random_state=random_state, 

34 max_leaf_nodes=max_leaf_nodes, 

35 min_impurity_decrease=min_impurity_decrease) 

36 

37 def fit(self, X, y, sample_weight=None, check_input=True): 

38 """ 

39 Replaces the string stored in criterion by an instance of a class. 

40 """ 

41 replace = None 

42 if isinstance(self.criterion, str): 

43 if self.criterion == 'mselin': 

44 from .piecewise_tree_regression_criterion_linear import ( # pylint: disable=E0611,C0415 

45 LinearRegressorCriterion) 

46 replace = self.criterion 

47 self.criterion = LinearRegressorCriterion( 

48 1 if len(y.shape) <= 1 else y.shape[1], X) 

49 elif self.criterion == "simple": 

50 from .piecewise_tree_regression_criterion_fast import ( # pylint: disable=E0611,C0415 

51 SimpleRegressorCriterionFast) 

52 replace = self.criterion 

53 self.criterion = SimpleRegressorCriterionFast( 

54 1 if len(y.shape) <= 1 else y.shape[1], X.shape[0]) 

55 else: 

56 replace = None 

57 

58 DecisionTreeRegressor.fit( 

59 self, X, y, 

60 sample_weight=sample_weight, 

61 check_input=check_input) 

62 

63 if replace: 

64 self.criterion = replace 

65 

66 if self.criterion == "mselin": 

67 self._fit_reglin(X, y, sample_weight) 

68 return self 

69 

70 def _mapping_train(self, X): 

71 tree = self.tree_ 

72 leaves = [i for i in range(len(tree.children_left)) 

73 if tree.children_left[i] <= i and tree.children_right[i] <= i] # pylint: disable=E1136 

74 dec_path = self.decision_path(X) 

75 association = numpy.zeros((X.shape[0],)) 

76 association[:] = -1 

77 mapping = {} 

78 ntree = 0 

79 for j in leaves: 

80 ind = dec_path[:, j] == 1 

81 ind = numpy.asarray(ind.todense()).flatten() 

82 if not numpy.any(ind): 

83 # No training example for this bucket. 

84 continue 

85 mapping[j] = ntree 

86 association[ind] = ntree 

87 ntree += 1 

88 return mapping 

89 

90 def predict_leaves(self, X): 

91 """ 

92 Returns the leave index for each observation of *X*. 

93 

94 :param X: array 

95 :return: array 

96 leaves index in ``self.leaves_index_`` 

97 """ 

98 # The creation of the sparse matrix could be avoided. 

99 leaves = self.decision_path(X) 

100 leaves = leaves[:, self.leaves_index_] 

101 mat = numpy.argmax(leaves, 1) 

102 res = numpy.asarray(mat).ravel() 

103 return res 

104 

105 def _fit_reglin(self, X, y, sample_weight): 

106 """ 

107 Fits linear regressions for all leaves. 

108 Sets attributes ``leaves_mapping_``, ``betas_``, ``leaves_index_``. 

109 The first attribute is a dictionary ``{leave: row}`` 

110 which maps a leave of the tree to the coefficients 

111 ``betas_[row, :]`` of a regression trained on all training 

112 points mapped a specific leave. ``leaves_index_`` keeps 

113 in memory a set of leaves. 

114 """ 

115 from .piecewise_tree_regression_criterion_linear import ( # pylint: disable=E0611,C0415 

116 LinearRegressorCriterion) 

117 

118 tree = self.tree_ 

119 self.leaves_index_ = [i for i in range(len(tree.children_left)) 

120 if tree.children_left[i] <= i and tree.children_right[i] <= i] # pylint: disable=E1136 

121 if tree.n_leaves != len(self.leaves_index_): 

122 raise RuntimeError( # pragma: no cover 

123 f"Unexpected number of leaves {tree.n_leaves} " 

124 f"!= {len(self.leaves_index_)}.") 

125 pred_leaves = self.predict_leaves(X) 

126 self.leaves_mapping_ = {k: i for i, k in enumerate(pred_leaves)} 

127 self.betas_ = numpy.empty((len(self.leaves_index_), X.shape[1] + 1)) 

128 for i, _ in enumerate(self.leaves_index_): 

129 ind = pred_leaves == i 

130 xs = X[ind, :].copy() 

131 ys = y[ind].astype(numpy.float64) 

132 if len(ys.shape) == 1: 

133 ys = ys[:, numpy.newaxis] 

134 ys = ys.copy() 

135 ws = sample_weight[ind].copy() if sample_weight else None 

136 dec = LinearRegressorCriterion.create(xs, ys, ws) 

137 dec.node_beta(self.betas_[i, :]) 

138 

139 def predict(self, X, check_input=True): 

140 """ 

141 Overloads method *predict*. Falls back into 

142 the predict from a decision tree is criterion is 

143 *mse*, *mae*, *simple*. Computes the predictions 

144 from linear regression if the criterion is *mselin*. 

145 """ 

146 if self.criterion == 'mselin': 

147 return self._predict_reglin(X, check_input=check_input) 

148 return DecisionTreeRegressor.predict(self, X, check_input=check_input) 

149 

150 def _predict_reglin(self, X, check_input=True): 

151 """ 

152 Computes the predictions with a linear regression 

153 fitted with the observations mapped to each leave 

154 of the tree. 

155 

156 :param X: array-like or sparse matrix of shape = [n_samples, n_features] 

157 The input samples. Internally, it will be converted to 

158 ``dtype=np.float32`` and if a sparse matrix is provided 

159 to a sparse ``csr_matrix``. 

160 :param check_input: boolean, (default=True) 

161 Allow to bypass several input checking. 

162 Don't use this parameter unless you know what you do. 

163 :return: y, array of shape = [n_samples] or [n_samples, n_outputs] 

164 The predicted classes, or the predict values. 

165 """ 

166 leaves = self.predict_leaves(X) 

167 pred = numpy.ones((X.shape[0], 1)) 

168 Xone = numpy.hstack([X, pred]) 

169 for i in range(0, X.shape[0]): 

170 li = leaves[i] 

171 pred[i] = numpy.dot(Xone[i, :], self.betas_[li, :]) 

172 return pred.ravel()