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
213
214
215
216
217
218
219
220
221
222
|
# Test methods with long descriptive names can omit docstrings
# pylint: disable=missing-docstring
import unittest
import numpy as np
from Orange.classification import (CN2Learner, CN2UnorderedLearner,
CN2SDLearner, CN2SDUnorderedLearner)
from Orange.classification.rules import (_RuleLearner, _RuleClassifier,
RuleHunter, Rule, EntropyEvaluator,
LaplaceAccuracyEvaluator,
WeightedRelativeAccuracyEvaluator,
argmaxrnd, hash_dist)
from Orange.data import Table
from Orange.data.filter import HasClass
from Orange.preprocess import Impute
class TestRuleInduction(unittest.TestCase):
def setUp(self):
self.titanic = Table('titanic')
self.iris = Table('iris')
def test_base_RuleLearner(self):
"""
Base rule induction learner test. To pass the test, all base
components are checked, including preprocessors, top-level
control procedure elements (covering algorithm, rule stopping,
data stopping), and bottom-level search procedure controller
(rule finder).
Every learner that extends _RuleLearner should override the fit
method. It should at this point not yet be available (exception
raised).
"""
base_rule_learner = _RuleLearner()
# test the number of default preprocessors
self.assertEqual(len(list(base_rule_learner.active_preprocessors)), 3)
# preprocessor types
preprocessor_types = [type(x) for x in base_rule_learner.active_preprocessors]
self.assertIn(HasClass, preprocessor_types)
self.assertIn(Impute, preprocessor_types)
# test find_rules
base_rule_learner.domain = self.iris.domain
base_rule_learner.find_rules(self.iris.X, self.iris.Y.astype(int),
None, None, [], self.iris.domain)
# test top-level control procedure elements
self.assertTrue(hasattr(base_rule_learner, "data_stopping"))
self.assertTrue(hasattr(base_rule_learner, "cover_and_remove"))
self.assertTrue(hasattr(base_rule_learner, "rule_stopping"))
# test exclusive covering algorithm
new_rule = Rule()
new_rule.covered_examples = np.array([True, False, True], dtype=bool)
new_rule.target_class = None
X, Y, W = base_rule_learner.exclusive_cover_and_remove(
self.iris.X[:3], self.iris.Y[:3], None, new_rule)
self.assertTrue(len(X) == len(Y) == 1)
# test rule finder
self.assertTrue(hasattr(base_rule_learner, "rule_finder"))
rule_finder = base_rule_learner.rule_finder
self.assertIsInstance(rule_finder, RuleHunter)
self.assertTrue(hasattr(rule_finder, "search_algorithm"))
self.assertTrue(hasattr(rule_finder, "search_strategy"))
self.assertTrue(hasattr(rule_finder, "quality_evaluator"))
self.assertTrue(hasattr(rule_finder, "complexity_evaluator"))
self.assertTrue(hasattr(rule_finder, "general_validator"))
self.assertTrue(hasattr(rule_finder, "significance_validator"))
def testBaseRuleClassifier(self):
"""
Every classifier that extends _RuleClassifier should override
the predict method. It should at this point not yet be available
(exception raised).
"""
base_rule_classifier = _RuleClassifier(domain=self.iris.domain)
self.assertRaises(NotImplementedError, base_rule_classifier.predict,
self.iris.X)
def testCN2Learner(self):
learner = CN2Learner()
# classic CN2 removes covered learning instances
self.assertTrue(learner.cover_and_remove ==
_RuleLearner.exclusive_cover_and_remove)
# Entropy measure is used to evaluate found hypotheses
self.assertTrue(type(learner.rule_finder.quality_evaluator) ==
EntropyEvaluator)
# test that the learning requirements are relaxed by default
self.assertTrue(learner.rule_finder.general_validator.max_rule_length >= 5)
self.assertTrue(learner.rule_finder.general_validator.min_covered_examples == 1)
classifier = learner(self.titanic)
self.assertEqual(classifier.original_domain, self.titanic.domain)
# all learning instances are covered when limitations do not
# impose rule length or minimum number of covered examples
num_covered = np.sum([rule.curr_class_dist
for rule in classifier.rule_list[:-1]])
self.assertEqual(num_covered, self.titanic.X.shape[0])
# prediction (matrix-wise, all testing instances at once)
# test that returned result is of correct size
predictions = classifier.predict(self.titanic.X)
self.assertEqual(len(predictions), self.titanic.X.shape[0])
def testCN2PrefersEquality(self):
learner = CN2Learner()
classifier = learner(self.titanic)
operators = [s.op for rule in classifier.rule_list for s in rule.selectors]
self.assertEqual(operators.count('!='), 4)
self.assertEqual(operators.count('=='), 23)
def testCN2RestrictEquality(self):
learner = CN2Learner(restrict_equality=True)
classifier = learner(self.titanic)
operators = [s.op for rule in classifier.rule_list for s in rule.selectors]
self.assertEqual(operators.count('!='), 0)
def testUnorderedCN2Learner(self):
learner = CN2UnorderedLearner()
# unordered CN2 removes covered learning instances
self.assertTrue(learner.cover_and_remove ==
_RuleLearner.exclusive_cover_and_remove)
# Laplace accuracy measure is used to evaluate found hypotheses
self.assertTrue(type(learner.rule_finder.quality_evaluator) ==
LaplaceAccuracyEvaluator)
# test that the learning requirements are relaxed by default
self.assertTrue(learner.rule_finder.general_validator.max_rule_length >= 5)
self.assertTrue(learner.rule_finder.general_validator.min_covered_examples == 1)
# by default, continuous variables are
# constrained by the learning algorithm
self.assertTrue(learner.rule_finder.search_strategy.constrain_continuous)
classifier = learner(self.iris)
self.assertEqual(classifier.original_domain, self.iris.domain)
# all learning instances should be covered given the parameters
for curr_class in range(len(self.iris.domain.class_var.values)):
target_covered = (np.sum([rule.curr_class_dist[rule.target_class]
for rule in classifier.rule_list
if rule.target_class == curr_class]))
self.assertEqual(target_covered, np.sum(self.iris.Y == curr_class))
# a custom example, test setting several parameters
learner = CN2UnorderedLearner()
learner.rule_finder.search_algorithm.beam_width = 5
learner.rule_finder.search_strategy.constrain_continuous = True
learner.rule_finder.general_validator.min_covered_examples = 10
learner.rule_finder.general_validator.max_rule_length = 2
learner.rule_finder.significance_validator.parent_alpha = 0.9
learner.rule_finder.significance_validator.default_alpha = 0.8
classifier = learner(self.iris)
# only the TRUE rule may exceed imposed limitations
for rule in classifier.rule_list[:-1]:
self.assertLessEqual(len(rule.selectors), 2)
self.assertGreaterEqual(np.max(rule.curr_class_dist), 10)
# prediction (matrix-wise, all testing instances at once)
# test that returned result is of correct size
predictions = classifier.predict(self.iris.X)
self.assertEqual(len(predictions), self.iris.X.shape[0])
def testOrderedCN2SDLearner(self):
learner = CN2SDLearner()
# Weighted relative accuracy measure is
# used to evaluate found hypotheses
self.assertTrue(type(learner.rule_finder.quality_evaluator) ==
WeightedRelativeAccuracyEvaluator)
# gamma parameter must be initialized and defined
self.assertTrue(hasattr(learner, "gamma"))
classifier = learner(self.titanic)
self.assertEqual(classifier.original_domain, self.titanic.domain)
# prediction (matrix-wise, all testing instances at once)
# test that returned result is of correct size
predictions = classifier.predict(self.titanic.X)
self.assertEqual(len(predictions), self.titanic.X.shape[0])
def testUnorderedCN2SDLearner(self):
learner = CN2SDUnorderedLearner()
learner.rule_finder.significance_validator.parent_alpha = 0.2
learner.rule_finder.significance_validator.default_alpha = 0.8
self.assertTrue(type(learner.rule_finder.quality_evaluator) ==
WeightedRelativeAccuracyEvaluator)
# gamma parameter must be initialized and defined
self.assertTrue(hasattr(learner, "gamma"))
classifier = learner(self.titanic)
self.assertEqual(classifier.original_domain, self.titanic.domain)
# prediction (matrix-wise, all testing instances at once)
# test that returned result is of correct size
predictions = classifier.predict(self.titanic.X)
self.assertEqual(len(predictions), self.titanic.X.shape[0])
def testArgMaxRnd(self):
temp = np.array([np.nan, 1, 2.3, 37, 37, 37, 1])
self.assertEqual(argmaxrnd(temp, hash_dist(np.array([3, 4]))), 5)
self.assertRaises(ValueError, argmaxrnd, np.ones((1, 1, 1)))
if __name__ == '__main__':
unittest.main()
|
