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
|
import unittest
import numpy as np
from Orange.data import Table, DiscreteVariable, Domain
from Orange.classification import LogisticRegressionLearner, TreeLearner
class TestModelMapping(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.iris = iris = Table("iris")
tables = []
ix = iris.X
y = np.hstack((np.zeros(50), np.ones(50)))
attrs = cls.iris.domain.attributes
classes = cls.iris.domain.class_var.values
for i, x in enumerate([ix[50:],
np.vstack((ix[:50], ix[100:])),
ix[:100]]):
class_var = DiscreteVariable(
"iris",
values=tuple(n for j, n in enumerate(classes) if j != i))
domain = Domain(attrs, class_var)
tables.append(Table.from_numpy(domain, x, y))
# pylint: disable=unbalanced-tuple-unpacking
cls.iris0, cls.iris1, cls.iris2 = tables
def test_larger_model(self):
# Train on all data, test on subset of values
# Probabilities should stay the same, but normalized
# Class predictions for existing classes should stay the same
def normalized(a):
n = np.sum(a, axis=1)
n[n == 0] = 1.0 / a.shape[1]
a[n == 0] = 1
return a / n[:, None]
for lrn in [TreeLearner, LogisticRegressionLearner]: # skl and non-skl
model = lrn()(self.iris)
val, prob = model(self.iris, model.ValueProbs)
val0, prob0 = model(self.iris0, model.ValueProbs)
vale = val[50:]
probe = normalized(prob[50:, 1:])
# No effect on class predictions
np.testing.assert_array_equal(
val0[vale != 0],
vale[vale != 0] - 1)
# Classes that to not exist are replaced with the most probable;
# don't use argmax because of possible ties
np.testing.assert_array_equal(
probe[vale == 0, val0[vale == 0].astype(int)],
np.max(probe[vale == 0], axis=1))
# Probabilities are not affected (but normalized)
np.testing.assert_almost_equal(prob0, probe)
# Same as above for other two classes ...
val1, prob1 = model(self.iris1, model.ValueProbs)
vale = np.hstack((val[:50], val[100:])).astype(float)
no1 = vale != 1
vale[vale == 2] -= 1
probe = np.vstack((prob[:50], prob[100:]))
probe = normalized(np.hstack((probe[:, :1], probe[:, 2:])))
np.testing.assert_array_equal(
val1[no1],
vale[no1])
np.testing.assert_array_equal(
probe[~no1, val1[~no1].astype(int)],
np.max(probe[~no1], axis=1))
np.testing.assert_almost_equal(prob1, probe)
val2, prob2 = model(self.iris2, model.ValueProbs)
vale = val[:100]
probe = normalized(prob[:100, :2])
np.testing.assert_array_equal(
val2[vale != 2],
vale[vale != 2])
np.testing.assert_array_equal(
probe[vale != 2, val2[vale != 2].astype(int)],
np.max(probe[vale != 2], axis=1))
np.testing.assert_almost_equal(prob2, probe)
def test_smaller_model(self):
for lrn in [LogisticRegressionLearner, TreeLearner]: # skl and non-skl
model = lrn()(self.iris0)
val0, prob0 = model(self.iris0, model.ValueProbs)
val, prob = model(self.iris, model.ValueProbs)
# Model can't predict class 0 in whole data
np.testing.assert_array_equal(val0, val[50:] - 1)
np.testing.assert_almost_equal(prob0, prob[50:, 1:])
# First 50 instances in whole data can be assigned anything 1 or 2
# and should not be nan
self.assertTrue(np.all((val[:50] == 1) + (val[:50] == 2)))
np.testing.assert_almost_equal(np.sum(prob, axis=1), 1)
np.testing.assert_almost_equal(prob[:, 0], 0)
model = lrn()(self.iris1)
val, prob = model(self.iris, model.ValueProbs)
val1, prob1 = model(self.iris1, model.ValueProbs)
np.testing.assert_array_equal(val1[:50], val[:50])
np.testing.assert_array_equal(val1[50:], val[100:] - 1)
np.testing.assert_almost_equal(prob1[:50, 0], prob[:50, 0])
np.testing.assert_almost_equal(prob1[:50, 1], prob[:50, 2])
np.testing.assert_almost_equal(prob1[50:, 0], prob[100:, 0])
np.testing.assert_almost_equal(prob1[50:, 1], prob[100:, 2])
self.assertTrue(np.all((val[50:100] == 0) + (val[50:100] == 2)))
np.testing.assert_almost_equal(np.sum(prob, axis=1), 1)
np.testing.assert_almost_equal(prob[:, 1], 0)
model = lrn()(self.iris2)
val, prob = model(self.iris, model.ValueProbs)
val2, prob2 = model(self.iris2, model.ValueProbs)
np.testing.assert_array_equal(val2, val[:100])
np.testing.assert_almost_equal(prob2, prob[:100, :2])
self.assertTrue(np.all((val[100:] == 0) + (val[100:] == 1)))
self.assertTrue(np.all(val[:50] < 2)) # also tests it's not nan
np.testing.assert_almost_equal(np.sum(prob, axis=1), 1)
np.testing.assert_almost_equal(prob[:, 2], 0)
def test_model_different(self):
def test_val_prob(val, prob):
np.testing.assert_almost_equal(np.sum(prob, axis=1), 1)
np.testing.assert_array_equal(
np.choose(val.astype(int), (prob[:, 0], prob[:, 1])),
np.max(prob, axis=1))
for lrn in [LogisticRegressionLearner, TreeLearner]: # skl and non-skl
model0 = lrn()(self.iris0)
valp0 = model0(self.iris0)
model1 = lrn()(self.iris1)
valp1 = model1(self.iris1)
model2 = lrn()(self.iris2)
valp2 = model2(self.iris2)
val1, prob1 = model0(self.iris1, model0.ValueProbs)
np.testing.assert_array_equal(val1[valp0 == 1], 1)
np.testing.assert_array_equal(prob1[valp0 == 1, 0], 0)
np.testing.assert_array_equal(prob1[valp0 == 1, 1], 1)
test_val_prob(val1, prob1)
val2, prob2 = model0(self.iris2, model0.ValueProbs)
np.testing.assert_array_equal(val2[valp0 == 1], 1)
np.testing.assert_array_equal(prob2[valp0 == 1, 0], 0)
np.testing.assert_array_equal(prob2[valp0 == 1, 1], 1)
np.testing.assert_almost_equal(np.sum(prob2, axis=1), 1)
test_val_prob(val2, prob2)
val0, prob0 = model1(self.iris0, model1.ValueProbs)
np.testing.assert_array_equal(val0[valp1 == 1], 1)
np.testing.assert_array_equal(prob0[valp1 == 1, 0], 0)
np.testing.assert_array_equal(prob0[valp1 == 1, 1], 1)
np.testing.assert_almost_equal(np.sum(prob0, axis=1), 1)
test_val_prob(val0, prob0)
val2, prob2 = model1(self.iris2, model1.ValueProbs)
np.testing.assert_array_equal(val2[valp1 == 0], 0)
np.testing.assert_array_equal(prob2[valp1 == 0, 0], 1)
np.testing.assert_array_equal(prob2[valp1 == 0, 1], 0)
np.testing.assert_almost_equal(np.sum(prob2, axis=1), 1)
test_val_prob(val2, prob2)
val0, prob0 = model2(self.iris0, model2.ValueProbs)
np.testing.assert_array_equal(val0[valp2 == 1], 0)
np.testing.assert_array_equal(prob0[valp2 == 1, 0], 1)
np.testing.assert_array_equal(prob0[valp2 == 1, 1], 0)
np.testing.assert_almost_equal(np.sum(prob0, axis=1), 1)
test_val_prob(val0, prob0)
val1, prob1 = model2(self.iris1, model2.ValueProbs)
np.testing.assert_array_equal(val1[valp2 == 0], 0)
np.testing.assert_array_equal(prob1[valp2 == 0, 0], 1)
np.testing.assert_array_equal(prob1[valp2 == 0, 1], 0)
np.testing.assert_almost_equal(np.sum(prob1, axis=1), 1)
test_val_prob(val1, prob1)
def test_no_common_values(self):
abc = DiscreteVariable("iris", values=tuple("abc"))
iris_abc = Table.from_numpy(
Domain(self.iris.domain.attributes, abc),
self.iris.X, self.iris.Y)
for lrn in [LogisticRegressionLearner,
TreeLearner]: # skl and non-skl
model = lrn()(self.iris)
val, prob = model(iris_abc, model.ValueProbs)
self.assertTrue(np.all(val >= 0))
self.assertTrue(np.all(val <= 2))
np.testing.assert_array_equal(prob, 1 / 3)
def test_sparse_matrix(self):
iris_sparse = self.iris.to_sparse()
for lrn in [LogisticRegressionLearner, TreeLearner]: # skl and non-skl
model = lrn()(iris_sparse)
pred = model(iris_sparse.X.tocsc())
self.assertTupleEqual((len(self.iris),), pred.shape)
pred = model(iris_sparse.X.tolil())
self.assertTupleEqual((len(self.iris),), pred.shape)
pred = model(iris_sparse.X.tocoo())
self.assertTupleEqual((len(self.iris),), pred.shape)
if __name__ == '__main__':
unittest.main()
|
