mlinsights/_unittests/ut_mlmodel/test_piecewise_decision_tree_experiment_linear.py at 3e3a39b4a892d8a788cf6bf7a3422ca47d616820 · sdpython/mlinsights · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
# -*- coding: utf-8 -*-
"""
@brief      test log(time=10s)
"""
import unittest
import numpy
from sklearn.tree._criterion import MSE  # pylint: disable=E0611
from sklearn.tree import DecisionTreeRegressor
from sklearn import datasets
from sklearn.model_selection import train_test_split
from pyquickhelper.pycode import ExtTestCase
from mlinsights.mlmodel.piecewise_tree_regression import PiecewiseTreeRegressor
from mlinsights.mlmodel._piecewise_tree_regression_common import (  # pylint: disable=E0611,E0401
    _test_criterion_init, _test_criterion_node_impurity,
    _test_criterion_node_impurity_children, _test_criterion_update,
    _test_criterion_node_value, _test_criterion_proxy_impurity_improvement,
    _test_criterion_impurity_improvement
)
from mlinsights.mlmodel.piecewise_tree_regression_criterion_linear import LinearRegressorCriterion  # pylint: disable=E0611, E0401


class TestPiecewiseDecisionTreeExperimentLinear(ExtTestCase):

    def test_criterions(self):
        X = numpy.array([[10., 12., 13.]]).T
        y = numpy.array([20., 22., 23.])
        c1 = MSE(1, X.shape[0])
        c2 = LinearRegressorCriterion(1, X)
        self.assertNotEmpty(c1)
        self.assertNotEmpty(c2)
        w = numpy.ones((y.shape[0],))
        self.assertEqual(w.sum(), X.shape[0])
        ind = numpy.arange(y.shape[0]).astype(numpy.int64)
        ys = y.astype(float).reshape((y.shape[0], 1))
        _test_criterion_init(c1, ys, w, 1., ind, 0, y.shape[0])
        _test_criterion_init(c2, ys, w, 1., ind, 0, y.shape[0])
        # https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/tree/_criterion.pyx#L886
        v1 = _test_criterion_node_value(c1)
        v2 = _test_criterion_node_value(c2)
        self.assertEqual(v1, v2)
        i1 = _test_criterion_node_impurity(c1)
        i2 = _test_criterion_node_impurity(c2)
        self.assertGreater(i1, i2)
        self.assertGreater(i2, 0)
        p1 = _test_criterion_proxy_impurity_improvement(c1)
        p2 = _test_criterion_proxy_impurity_improvement(c2)
        self.assertTrue(numpy.isnan(p1), numpy.isnan(p2))

        X = numpy.array([[1., 2., 3.]]).T
        y = numpy.array([1., 2., 3.])
        c1 = MSE(1, X.shape[0])
        c2 = LinearRegressorCriterion(1, X)
        w = numpy.ones((y.shape[0],))
        ind = numpy.arange(y.shape[0]).astype(numpy.int64)
        ys = y.astype(float).reshape((y.shape[0], 1))
        _test_criterion_init(c1, ys, w, 1., ind, 0, y.shape[0])
        _test_criterion_init(c2, ys, w, 1., ind, 0, y.shape[0])
        i1 = _test_criterion_node_impurity(c1)
        i2 = _test_criterion_node_impurity(c2)
        self.assertGreater(i1, i2)
        v1 = _test_criterion_node_value(c1)
        v2 = _test_criterion_node_value(c2)
        self.assertEqual(v1, v2)
        p1 = _test_criterion_proxy_impurity_improvement(c1)
        p2 = _test_criterion_proxy_impurity_improvement(c2)
        self.assertTrue(numpy.isnan(p1), numpy.isnan(p2))

        X = numpy.array([[1., 2., 10., 11.]]).T
        y = numpy.array([0.9, 1.1, 1.9, 2.1])
        c1 = MSE(1, X.shape[0])
        c2 = LinearRegressorCriterion(1, X)
        w = numpy.ones((y.shape[0],))
        ind = numpy.arange(y.shape[0]).astype(numpy.int64)
        ys = y.astype(float).reshape((y.shape[0], 1))
        _test_criterion_init(c1, ys, w, 1., ind, 0, y.shape[0])
        _test_criterion_init(c2, ys, w, 1., ind, 0, y.shape[0])
        i1 = _test_criterion_node_impurity(c1)
        i2 = _test_criterion_node_impurity(c2)
        self.assertGreater(i1, i2)
        v1 = _test_criterion_node_value(c1)
        v2 = _test_criterion_node_value(c2)
        self.assertEqual(v1, v2)
        p1 = _test_criterion_proxy_impurity_improvement(c1)
        p2 = _test_criterion_proxy_impurity_improvement(c2)
        self.assertTrue(numpy.isnan(p1), numpy.isnan(p2))

        X = numpy.array([[1., 2., 10., 11.]]).T
        y = numpy.array([0.9, 1.1, 1.9, 2.1])
        c1 = MSE(1, X.shape[0])
        c2 = LinearRegressorCriterion(1, X)
        w = numpy.ones((y.shape[0],))
        ind = numpy.array([0, 3, 2, 1], dtype=ind.dtype)
        ys = y.astype(float).reshape((y.shape[0], 1))
        _test_criterion_init(c1, ys, w, 1., ind, 1, y.shape[0])
        _test_criterion_init(c2, ys, w, 1., ind, 1, y.shape[0])
        i1 = _test_criterion_node_impurity(c1)
        i2 = _test_criterion_node_impurity(c2)
        self.assertGreater(i1, i2)
        v1 = _test_criterion_node_value(c1)
        v2 = _test_criterion_node_value(c2)
        self.assertEqual(v1, v2)
        p1 = _test_criterion_proxy_impurity_improvement(c1)
        p2 = _test_criterion_proxy_impurity_improvement(c2)
        self.assertTrue(numpy.isnan(p1), numpy.isnan(p2))

        for i in range(2, 4):
            _test_criterion_update(c1, i)
            _test_criterion_update(c2, i)
            left1, right1 = _test_criterion_node_impurity_children(c1)
            left2, right2 = _test_criterion_node_impurity_children(c2)
            self.assertGreater(left1, left2)
            self.assertGreater(right1, right2)
            v1 = _test_criterion_node_value(c1)
            v2 = _test_criterion_node_value(c2)
            self.assertEqual(v1, v2)
            try:
                # scikit-learn >= 0.24
                p1 = _test_criterion_impurity_improvement(
                    c1, 0., left1, right1)
                p2 = _test_criterion_impurity_improvement(
                    c2, 0., left2, right2)
            except TypeError:
                # scikit-learn < 0.23
                p1 = _test_criterion_impurity_improvement(c1, 0.)
                p2 = _test_criterion_impurity_improvement(c2, 0.)
            self.assertGreater(p1, p2 - 1.)

            dest = numpy.empty((2, ))
            c2.node_beta(dest)
            self.assertGreater(dest[0], 0)
            self.assertGreater(dest[1], 0)

    def test_criterions_check_value(self):
        X = numpy.array([[10., 12., 13.]]).T
        y = numpy.array([[20., 22., 23.]]).T
        c2 = LinearRegressorCriterion.create(X, y)
        coef = numpy.empty((3, ))
        c2.node_beta(coef)
        self.assertEqual(coef[:2], numpy.array([1, 10]))

    def test_decision_tree_criterion(self):
        X = numpy.array([[1., 2., 10., 11.]]).T
        y = numpy.array([0.9, 1.1, 1.9, 2.1])
        clr1 = DecisionTreeRegressor(max_depth=1)
        clr1.fit(X, y)
        p1 = clr1.predict(X)

        crit = LinearRegressorCriterion(1, X)
        clr2 = DecisionTreeRegressor(criterion=crit, max_depth=1)
        clr2.fit(X, y)
        p2 = clr2.predict(X)
        self.assertEqual(p1, p2)
        self.assertEqual(clr1.tree_.node_count, clr2.tree_.node_count)

    def test_decision_tree_criterion_iris(self):
        iris = datasets.load_iris()
        X, y = iris.data, iris.target
        clr1 = DecisionTreeRegressor()
        clr1.fit(X, y)
        p1 = clr1.predict(X)
        clr2 = DecisionTreeRegressor(criterion=LinearRegressorCriterion(1, X))
        clr2.fit(X, y)
        p2 = clr2.predict(X)
        self.assertEqual(p1.shape, p2.shape)

    def test_decision_tree_criterion_iris_dtc(self):
        iris = datasets.load_iris()
        X, y = iris.data, iris.target
        clr1 = DecisionTreeRegressor()
        clr1.fit(X, y)
        p1 = clr1.predict(X)
        clr2 = PiecewiseTreeRegressor(criterion='mselin')
        clr2.fit(X, y)
        p2 = clr2.predict(X)
        self.assertEqual(p1.shape, p2.shape)
        self.assertTrue(hasattr(clr2, 'betas_'))
        self.assertTrue(hasattr(clr2, 'leaves_mapping_'))
        self.assertEqual(len(clr2.leaves_index_), clr2.tree_.n_leaves)
        self.assertEqual(len(clr2.leaves_mapping_), clr2.tree_.n_leaves)
        self.assertEqual(clr2.betas_.shape[1], X.shape[1] + 1)
        self.assertEqual(clr2.betas_.shape[0], clr2.tree_.n_leaves)
        sc1 = clr1.score(X, y)
        sc2 = clr2.score(X, y)
        self.assertGreater(sc1, sc2)
        mp = clr2._mapping_train(X)  # pylint: disable=W0212
        self.assertIsInstance(mp, dict)
        self.assertGreater(len(mp), 2)

    def test_decision_tree_criterion_iris_dtc_traintest(self):
        iris = datasets.load_iris()
        X, y = iris.data, iris.target
        X_train, X_test, y_train, y_test = train_test_split(X, y)
        clr1 = DecisionTreeRegressor()
        clr1.fit(X_train, y_train)
        p1 = clr1.predict(X_train)
        clr2 = PiecewiseTreeRegressor(criterion='mselin')
        clr2.fit(X_train, y_train)
        p2 = clr2.predict(X_train)
        self.assertEqual(p1.shape, p2.shape)
        self.assertTrue(hasattr(clr2, 'betas_'))
        self.assertTrue(hasattr(clr2, 'leaves_mapping_'))
        self.assertEqual(len(clr2.leaves_index_), clr2.tree_.n_leaves)
        self.assertEqual(len(clr2.leaves_mapping_), clr2.tree_.n_leaves)
        self.assertEqual(clr2.betas_.shape[1], X.shape[1] + 1)
        self.assertEqual(clr2.betas_.shape[0], clr2.tree_.n_leaves)
        sc1 = clr1.score(X_test, y_test)
        sc2 = clr2.score(X_test, y_test)
        self.assertGreater(abs(sc1 - sc2), -0.1)


if __name__ == "__main__":
    unittest.main()