def test_singleton_batch(self):
"""
In the case of a batch size of 1, y_true has shape (number_of_classes,) and accuracy should be correctly computed.
"""
y_true = np.ones((10,))
y_true_compare = np.ones((1, 10))
y_prob = -np.ones((1, 10))
assert_equals(accuracy.hamming_accuracy_from_blob(y_true, y_prob),
accuracy.hamming_accuracy_from_blob(y_true_compare, y_prob))
def test_accuracy_0(self):
"""
Extreme case with accuracy 0 and no threshold
"""
for n in range(2, 100):
y_true = np.ones((n, n))
y_prob = -np.ones((n, n)) / n
assert_equals(accuracy.hamming_accuracy_from_blob(y_true, y_prob), 0)
def test_accuracy_1_with_threshold_0(self):
"""
Accuracy 0 if all values under threshold
"""
threshold = 1
for n in range(2, 100):
y_true = np.ones((n, n))
y_prob = np.ones((n, n)) / n
assert_equals(accuracy.hamming_accuracy_from_blob(y_true, y_prob, threshold), 0)
def test_extra_dimension(self):
"""
If y_true has extra dimensions, we use a regular compact matricial form in 2D.
"""
y_true = np.array([[[[1]],
[[0]],
[[0]]],
[[[0]],
[[1]],
[[0]]],
[[[0]],
[[0]],
[[1]]], ])
y_true_compare = np.identity(3)
y_prob = np.array([[+1, -1, -1],
[-1, +1, -1],
[-1, -1, +1]])
assert_equals(accuracy.hamming_accuracy_from_blob(y_true, y_prob),
accuracy.hamming_accuracy_from_blob(y_true_compare, y_prob))
