def __call__(self, y: TensorLike) -> TensorLike:
self._assert_cut_contains_only_one_tensor(y)
if len(B.int_shape(y)) == 2:
return sum([y[:, ch] for ch in self._channels])
elif len(B.int_shape(y)) == 3:
return sum([self._agg_fn(y[:, :, ch]) for ch in self._channel])
elif len(B.int_shape(y)) == 4:
if self._channel_ax == 1:
return sum([self._agg_fn(y[:, ch]) for ch in self._channels])
elif self._channel_ax == 3:
return sum(
[self._agg_fn(y[:, :, :, ch]) for ch in self._channels])
else:
raise ValueError(
'Unsupported channel axis for convolutional layer: {}'.
format(self._channel_ax))
else:
raise QoiCutSupportError(
'Unsupported tensor rank for `InternalChannelQoI`: {}'.format(
len(B.int_shape(y))))
def test_gaussian(self):
doi = GaussianDoi(var=1., resolution=10)
res = doi(self.z)
self.assertEqual(len(res), 10,
'GaussianDoi should return `resolution` points')
self.assertEqual(B.int_shape(res[0]), B.int_shape(self.z))
def __call__(self, x: TensorLike) -> TensorLike:
self._assert_cut_contains_only_one_tensor(x)
if self.activation is not None:
if isinstance(self.activation, str):
self.activation = self.activation.lower()
if self.activation in ['sigmoid', 'softmax']:
x = getattr(B, self.activation)(x)
else:
raise NotImplementedError(
'This activation function is not currently supported '
'by the backend')
else:
x = self.activation(x)
# TODO(klas): is the `clone` necessary here? Not sure why it was
# included.
mask = B.sign(B.clone(x) - self.threshold)
if self.low_minus_high:
mask = -mask
non_batch_dimensions = tuple(range(len(B.int_shape(x)))[1:])
return B.sum(mask * x, axis=non_batch_dimensions)
def test_comparative(self):
qoi = ComparativeQoI(1, 0)
res = qoi(self.y)
self.assertEqual(B.int_shape(res), (2, ),
'Should return one scalar per row in the batch')
self.assertTrue(np.allclose(B.as_array(res), np.array([1., -1.])))
def test_class(self):
qoi = ClassQoI(1)
res = qoi(self.y)
self.assertEqual(B.int_shape(res), (2, ),
'Should return one scalar per row in the batch')
self.assertTrue(np.allclose(B.as_array(res), np.array([2., -1.])))
def test_internal_channel_axis3(self):
qoi = InternalChannelQoI(1, channel_axis=3)
res = qoi(self.z)
self.assertEqual(B.int_shape(res), (2, ),
'Should return one scalar per row in the batch')
self.assertTrue(np.allclose(B.as_array(res), np.array([21., 14.])))
def test_internal_channel_1d(self):
qoi = InternalChannelQoI(2)
res = qoi(self.y)
self.assertEqual(B.int_shape(res), (2, ),
'Should return one scalar per row in the batch')
self.assertTrue(np.allclose(B.as_array(res), np.array([3., -2.])))
def test_threshold_low_minus_high(self):
qoi = ThresholdQoI(1.5, low_minus_high=True)
res = qoi(self.y)
self.assertEqual(B.int_shape(res), (2, ),
'Should return one scalar per row in the batch')
self.assertTrue(np.allclose(B.as_array(res), np.array([-4., -3.])))
def test_lambda(self):
qoi = LambdaQoI(lambda y: y[:, 0] + y[:, 1])
res = qoi(self.y)
self.assertEqual(B.int_shape(res), (2, ),
'Should return one scalar per row in the batch')
self.assertTrue(np.allclose(B.as_array(res), np.array([3., -1.])))
def test_gaussian_non_tensor(self):
doi = GaussianDoi(var=1., resolution=10)
res = doi(B.as_array(self.z))
self.assertEqual(len(res), 10,
'GaussianDoi should return `resolution` points')
self.assertEqual(res[0].shape, B.int_shape(self.z))
def test_threshold_activation(self):
qoi = ThresholdQoI(0.75, activation='sigmoid')
res = qoi(self.y)
self.assertEqual(B.int_shape(res), (2, ),
'Should return one scalar per row in the batch')
self.assertTrue(
np.allclose(B.as_array(res), np.array([1.1023126, -0.8881443])))
def test_linear(self):
doi = LinearDoi(baseline=np.ones(B.int_shape(self.z)), resolution=21)
res = doi(self.z)
self.assertEqual(len(res), 21,
'LinearDoi should return `resolution` points')
self.assertTrue(np.array_equal(B.as_array(res[0]), B.as_array(self.z)),
'First point should be the original point')
self.assertTrue(np.all(B.as_array(res[-1]) == 1.),
'Last point should be baseline')
self.assertTrue(
np.allclose(B.as_array(res[-2]),
np.array([[1., 1.05, 1.1], [0.95, 0.9, 0.85]])),
'Intermediate points should interpolate from baseline')
