/usr/lib/python2.7/dist-packages/pebl/test/test_cpd.py is in python-pebl 1.0.2-4.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 | from numpy import array, allclose
from pebl import data, cpd
from pebl.test import testfile
def test_cextension():
try:
from pebl import _cpd
except:
assert False
class TestCPD_Py:
"""
Deriving loglikelihood manually
-------------------------------
Below is the derived calculation for the loglikelihood of the parentset for
node 0. Calculation done according to the g function from Cooper and
Herskovits. This test is done with binary varaibles because doing more on paper
is tedious. There are other tests that check for correct loglikelihood with
more complicated data.
data: 0110 parentset: {1,2,3} --> {0}
1001
1110
1110
0011
ri = child.arity = 2
parent config - (Nij+ri-1)! - Pi[Nijk!]
-------------------------------------------
000 (0+2-1)! 0!0!
001 (1+2-1)! 0!1!
010 (0+2-1)! 0!0!
011 (1+2-1)! 1!0!
100 (0+2-1)! 0!0!
101 (0+2-1)! 0!0!
110 (3+2-1)! 1!2!
111 (0+2-1)! 0!0!
likelihood = Pi[[(ri-1)!/(Nij+ri-1)!] Pi[Nijk])
= 1!0!0!/1! x 1!0!1!/2! x 1!0!0!/1! x
1!1!0!/2! x 1!1!2!/4! x 1!0!0!/1!
= 1 x 1/2 x 1 x
1/2 x 1/12 x 1
= 1/48
loglikelihood = ln(1/48) = -3.87120101107
"""
cpdtype = cpd.MultinomialCPD_Py
def setUp(self):
self.data = data.Dataset(array([[0, 1, 1, 0],
[1, 0, 0, 1],
[1, 1, 1, 0],
[1, 1, 1, 0],
[0, 0, 1, 1]]))
for v in self.data.variables:
v.arity = 2
self.cpd = self.cpdtype(self.data)
def test_lnfactorial_cache(self):
expected = array([ 0. , 0. , 0.69314718, 1.79175947,
3.17805383, 4.78749174, 6.57925121, 8.52516136,
10.6046029, 12.80182748, 15.10441257, 17.50230785,
19.9872145, 22.55216385, 25.19122118, 27.89927138,
30.67186011])
assert allclose(self.cpd.lnfactorial_cache, expected)
def test_offsets(self):
assert (self.cpd.offsets == array([0,1,2,4])).all()
def test_counts(self):
expected = array([[0, 0, 0],
[0, 0, 0],
[0, 0, 0],
[1, 2, 3],
[0, 1, 1],
[0, 0, 0],
[1, 0, 1],
[0, 0, 0]])
assert (self.cpd.counts == expected).all()
def loglikelihood(self):
assert allclose(self.cpd.loglikelihood(), -3.87120101091)
def test_replace1_loglikelihood(self):
# Do a noop replace.
self.cpd.replace_data(array([0,1,1,0]), array([0,1,1,0]))
assert allclose(self.cpd.loglikelihood(), -3.87120101091)
def test_replace1_counts(self):
self.cpd.replace_data(array([0,1,1,0]), array([0,1,1,0]))
expected = array([[0, 0, 0],
[0, 0, 0],
[0, 0, 0],
[1, 2, 3],
[0, 1, 1],
[0, 0, 0],
[1, 0, 1],
[0, 0, 0]])
assert (self.cpd.counts == expected).all()
def test_replace2_loglikelihood(self):
self.cpd.replace_data(self.data.observations[0], array([1,1,1,0]))
assert allclose(self.cpd.loglikelihood(), -2.77258872224)
def test_replace2_counts(self):
self.cpd.replace_data(self.data.observations[0], array([1,1,1,0]))
expected = array([[0, 0, 0],
[0, 0, 0],
[0, 0, 0],
[0, 3, 3],
[0, 1, 1],
[0, 0, 0],
[1, 0, 1],
[0, 0, 0]])
assert (self.cpd.counts == expected).all()
def test_undo_loglikelihood(self):
self.cpd.replace_data(self.data.observations[0], array([1,1,1,0]))
self.cpd.replace_data(array([1,1,1,0]),array([0,1,1,0]))
assert allclose(self.cpd.loglikelihood(), -3.87120101091)
def test_undo_counts(self):
self.cpd.replace_data(self.data.observations[0], array([1,1,1,0]))
self.cpd.replace_data(array([1,1,1,0]),array([0,1,1,0]))
expected = array([[0, 0, 0],
[0, 0, 0],
[0, 0, 0],
[1, 2, 3],
[0, 1, 1],
[0, 0, 0],
[1, 0, 1],
[0, 0, 0]])
assert (self.cpd.counts == expected).all()
def test_replace_with_ndarray(self):
self.cpd.replace_data(array([0,1,1,0]), array([1,1,1,0]))
assert allclose(self.cpd.loglikelihood(), -2.77258872224)
class TestCPD_C(TestCPD_Py):
cpdtype = cpd.MultinomialCPD_C
# The C version doesn't expose all datastructures to python
def test_lnfactorial_cache(self): pass
def test_offsets(self): pass
def test_counts(self): pass
def test_replace1_counts(self): pass
def test_replace2_counts(self): pass
def test_undo_counts(self): pass
class TestCPD2_Py:
"""
Can we properly handle nodes with no parents?
----------------------------------------------
With data=[1,0,1,1,0] for a node with no parents:
ri = child.arity = 2
parent config (Nij+ri-1)! Pi[Nijk!]
-------------------------------------------
null set (5+2-1)! 3!2!
likelihood = Pi[[(ri-1)!/(Nij+ri-1)!] Pi[Nijk])
= 1!3!2!/6!
= 12/720 = 1/60
loglikelihood = ln(1/60)
= -4.09434456
"""
cpdtype = cpd.MultinomialCPD_Py
def setUp(self):
self.data = data.Dataset(array([[1],
[0],
[1],
[1],
[0]]))
self.data.variables[0].arity = 2
self.cpd = self.cpdtype(self.data)
def test_offsets(self):
assert (self.cpd.offsets == array([0])).all()
def test_counts(self):
assert (self.cpd.counts == array([[2,3,5]])).all()
def test_loglikelihood(self):
assert allclose(self.cpd.loglikelihood(), -4.09434456)
class TestCPD2_C(TestCPD2_Py):
cpdtype = cpd.MultinomialCPD_C
# The C version doesn't expose all datastructures to python
def test_offsets(self): pass
def test_counts(self): pass
class TestMultinomialCPD_C:
def setUp(self):
self.data = data.fromfile(testfile("greedytest1-200.txt"))
def test_cpt_reuse(self):
# check that we don't have SegFault or BusError
# instead of freeing memory, _cpd will reuse it
for i in xrange(10000):
c = cpd.MultinomialCPD_C(self.data)
c.loglikelihood()
del c
def test_cpt_ceate_delete(self):
# "del c" will reuse memory while "del c2" will free it
for i in xrange(10000):
c = cpd.MultinomialCPD_C(self.data)
c2 = cpd.MultinomialCPD_C(self.data)
c.loglikelihood()
c2.loglikelihood()
del c
del c2
|