def test_weighted_mean_semantics(self):
inputs = np.array([1, 2, 3], dtype=np.float32)
weights1 = np.array([1, 1, 1], dtype=np.float32)
layer = metrics._WeightedMean()
full_signature = (signature(inputs), signature(weights1))
layer.init(full_signature)
mean1 = layer((inputs, weights1))
np.testing.assert_allclose(mean1, 2.0)
weights2 = np.array([0, 0, 1], dtype=np.float32)
mean2 = layer((inputs, weights2))
np.testing.assert_allclose(mean2, 3.0)
weights3 = np.array([1, 0, 0], dtype=np.float32)
mean3 = layer((inputs, weights3))
np.testing.assert_allclose(mean3, 1.0)
def test_weighted_mean_semantics(self):
layer = metrics._WeightedMean()
sample_input = np.ones((3, ))
sample_weights = np.ones((3, ))
layer.init(shapes.signature([sample_input, sample_weights]))
x = np.array([1., 2., 3.])
weights = np.array([1., 1., 1.])
mean = layer((x, weights))
np.testing.assert_allclose(mean, 2.)
weights = np.array([0., 0., 1.])
mean = layer((x, weights))
np.testing.assert_allclose(mean, 3.)
weights = np.array([1., 0., 0.])
mean = layer((x, weights))
np.testing.assert_allclose(mean, 1.)
def test_weighted_mean_shape(self):
input_signature = (ShapeDtype(
(29, 4, 4, 20)), ShapeDtype((29, 4, 4, 20)))
result_shape = base.check_shape_agreement(metrics._WeightedMean(),
input_signature)
self.assertEqual(result_shape, ())
def test_weighted_mean_shape(self):
layer = metrics._WeightedMean()
xs = [np.ones((9, 4, 4, 20)), np.ones((9, 4, 4, 20))]
y = layer(xs)
self.assertEqual(y.shape, ())