def test_happy_path_single(self):
softmax_values = np.array([0.1, 0.8, 0.08, 0.02])
softmax_values = np.expand_dims(softmax_values, 0)
prediction, entropies = SoftmaxEntropy.calculate(softmax_values)
self.assertEqual(1, prediction[0])
# https://www.wolframalpha.com/input/?i=-+%280.1*log2%280.1%29+%2B+0.8 [url continued in line below]
# *log2%280.8%29%2B0.08*log2%280.08%29%2B0.02*log2%280.02%29%29
self.assertAlmostEqual(0.9941209043760985826573513724, entropies[0])
def test_duplicate_non_winner(self):
softmax_values = np.array([[0.1, 0.8, 0.05, 0.05],
[0.2, 0.09, 0.7, 0.01]])
prediction, softmax = SoftmaxEntropy.calculate(softmax_values)
self.assertEqual((2, ), prediction.shape)
self.assertEqual((2, ), softmax.shape)
self.assertEqual(1, prediction[0])
# https://www.wolframalpha.com/input/?i=-+%280.1*log2%280.1%29+%2B+0.08 [url continued in line below]
# *log2%280.08%29%2B0.5*log2%280.5%29%2B0.05*log2%280.05%29%29
self.assertAlmostEqual(1.0219280948873623, softmax[0])
self.assertEqual(2, prediction[1])
# Calculation: See test above
self.assertAlmostEqual(1.203679208805967594, softmax[1])
def test_duplicate_winner(self):
softmax_values = np.array([[0.4, 0.4, 0.1, 0.1],
[0.2, 0.09, 0.7, 0.01]])
prediction, softmax = SoftmaxEntropy.calculate(softmax_values)
self.assertEqual((2, ), prediction.shape)
self.assertEqual((2, ), softmax.shape)
self.assertTrue(
0 == prediction[0] or 1 == prediction[0],
"Prediction must be index 0 or 1, but was {0}".format(
prediction[0]),
)
# https://www.wolframalpha.com/input/?i=-+%280.4*log2%280.4%29+%2B+0.4 [url continued in line below]
# *log2%280.4%29%2B0.1*log2%280.1%29%2B0.1*log2%280.1%29%29
self.assertAlmostEqual(1.7219280948873623, softmax[0])
self.assertEqual(2, prediction[1])
# Calculation: See test above
self.assertAlmostEqual(1.203679208805967594, softmax[1])
def test_happy_path_batch(self):
softmax_values = np.array([[0.1, 0.8, 0.08, 0.02],
[0.1, 0.8, 0.08, 0.02],
[0.2, 0.09, 0.7, 0.01]])
prediction, softmax = SoftmaxEntropy.calculate(softmax_values)
self.assertEqual((3, ), prediction.shape)
self.assertEqual((3, ), softmax.shape)
self.assertEqual(1, prediction[0])
# Calc: See test above
self.assertAlmostEqual(0.9941209043760985826573513724, softmax[0])
self.assertEqual(1, prediction[1])
# Calc: See test above
self.assertAlmostEqual(0.9941209043760985826573513724, softmax[1])
self.assertEqual(2, prediction[2])
# https://www.wolframalpha.com/input/?i=-+%280.2*log2%280.2%29+%2B+0.09 [url continued in line below]
# *log2%280.09%29%2B0.7*log2%280.7%29%2B0.01*log2%280.01%29%29
self.assertAlmostEqual(1.203679208805967594, softmax[2])
def test_problem_type(self):
self.assertEqual(SoftmaxEntropy.problem_type(),
ProblemType.CLASSIFICATION)
def test_samples_type_declaration(self):
self.assertFalse(SoftmaxEntropy.takes_samples())
def test_is_confidence(self):
self.assertFalse(SoftmaxEntropy.is_confidence())
self.assertFalse(SoftmaxEntropy().is_confidence())