def test_relu_deeplift_exact_match_wo_mutliplying_by_inputs(self) -> None:
x1 = torch.tensor([1.0])
x2 = torch.tensor([2.0])
inputs = (x1, x2)
model = ReLUDeepLiftModel()
dl = DeepLift(model, multiply_by_inputs=False)
attributions = dl.attribute(inputs)
self.assertEqual(attributions[0][0], 2.0)
self.assertEqual(attributions[1][0], 0.5)
def test_relu_linear_deeplift_compare_inplace(self) -> None:
model1 = ReLULinearDeepLiftModel(inplace=True)
x1 = torch.tensor([[-10.0, 1.0, -5.0], [2.0, 3.0, 4.0]], requires_grad=True)
x2 = torch.tensor([[3.0, 3.0, 1.0], [2.3, 5.0, 4.0]], requires_grad=True)
inputs = (x1, x2)
attributions1 = DeepLift(model1).attribute(inputs)
model2 = ReLULinearDeepLiftModel()
attributions2 = DeepLift(model2).attribute(inputs)
assertTensorAlmostEqual(self, attributions1[0], attributions2[0])
assertTensorAlmostEqual(self, attributions1[1], attributions2[1])
def test_relu_deeplift_with_hypothetical_contrib_func(self) -> None:
model = Conv1dDeepLiftModel()
rand_seq_data = torch.abs(torch.randn(2, 4, 1000))
rand_seq_ref = torch.abs(torch.randn(2, 4, 1000))
dls = DeepLift(model)
attr = dls.attribute(
rand_seq_data,
rand_seq_ref,
custom_attribution_func=_hypothetical_contrib_func,
target=(1, 0),
)
self.assertEqual(attr.shape, rand_seq_data.shape)
def test_lin_maxpool_lin_classification(self) -> None:
inputs = torch.ones(2, 4)
baselines = torch.tensor([[1, 2, 3, 9], [4, 8, 6, 7]]).float()
model = LinearMaxPoolLinearModel()
dl = DeepLift(model)
attrs, delta = dl.attribute(
inputs, baselines, target=0, return_convergence_delta=True
)
expected = [[0.0, 0.0, 0.0, -8.0], [0.0, -7.0, 0.0, 0.0]]
expected_delta = [0.0, 0.0]
assertArraysAlmostEqual(attrs.detach().numpy(), expected)
assertArraysAlmostEqual(delta.detach().numpy(), expected_delta)
def test_lin_maxpool_lin_classification(self) -> None:
inputs = torch.ones(2, 4)
baselines = torch.tensor([[1, 2, 3, 9], [4, 8, 6, 7]]).float()
model = LinearMaxPoolLinearModel()
dl = DeepLift(model)
attrs, delta = dl.attribute(inputs,
baselines,
target=0,
return_convergence_delta=True)
expected = torch.Tensor([[0.0, 0.0, 0.0, -8.0], [0.0, -7.0, 0.0, 0.0]])
expected_delta = torch.Tensor([0.0, 0.0])
assertTensorAlmostEqual(self, attrs, expected, 0.0001)
assertTensorAlmostEqual(self, delta, expected_delta, 0.0001)
def test_relu_deeplift_exact_match(self) -> None:
x1 = torch.tensor([1.0], requires_grad=True)
x2 = torch.tensor([2.0], requires_grad=True)
b1 = torch.tensor([0.0], requires_grad=True)
b2 = torch.tensor([0.0], requires_grad=True)
inputs = (x1, x2)
baselines = (b1, b2)
model = ReLUDeepLiftModel()
dl = DeepLift(model)
attributions, delta = dl.attribute(
inputs, baselines, return_convergence_delta=True
)
self.assertEqual(attributions[0][0], 2.0)
self.assertEqual(attributions[1][0], 1.0)
self.assertEqual(delta[0], 0.0)
def test_convnet_with_maxpool2d(self):
input = 100 * torch.randn(2, 1, 10, 10, requires_grad=True)
baseline = 20 * torch.randn(2, 1, 10, 10)
model = BasicModel_ConvNet()
dl = DeepLift(model)
self.softmax_classification(model, dl, input, baseline)
def test_convnet_with_maxpool1d_large_baselines(self) -> None:
input = 100 * torch.randn(2, 1, 10, requires_grad=True)
baseline = 500 * torch.randn(2, 1, 10)
model = BasicModel_ConvNet_MaxPool1d()
dl = DeepLift(model)
self.softmax_classification(model, dl, input, baseline, torch.tensor(2))
def test_relu_linear_deeplift_batch(self) -> None:
model = ReLULinearDeepLiftModel(inplace=True)
x1 = torch.tensor([[-10.0, 1.0, -5.0], [2.0, 3.0, 4.0]], requires_grad=True)
x2 = torch.tensor([[3.0, 3.0, 1.0], [2.3, 5.0, 4.0]], requires_grad=True)
inputs = (x1, x2)
baselines = (torch.zeros(1, 3), torch.rand(1, 3) * 0.001)
# expected = [[[0.0, 0.0]], [[6.0, 2.0]]]
self._deeplift_assert(model, DeepLift(model), inputs, baselines)
def test_softmax_classification_zero_baseline(self) -> None:
num_in = 20
input = torch.arange(0.0, num_in * 1.0, requires_grad=True).unsqueeze(0)
baselines = 0.0
model = SoftmaxDeepLiftModel(num_in, 20, 10)
dl = DeepLift(model)
self.softmax_classification(model, dl, input, baselines, torch.tensor(2))
def test_softmax_classification_batch_zero_baseline(self):
num_in = 40
input = torch.arange(0.0, num_in * 3.0, requires_grad=True).reshape(3, num_in)
baselines = 0 * input
model = SoftmaxDeepLiftModel(num_in, 20, 10)
dl = DeepLift(model)
self.softmax_classification(model, dl, input, baselines, torch.tensor(2))
def test_sigmoid_classification(self):
num_in = 20
input = torch.arange(0.0, num_in * 1.0, requires_grad=True).unsqueeze(0)
baseline = 0 * input
target = torch.tensor(0)
# TODO add test cases for multiple different layers
model = SigmoidDeepLiftModel(num_in, 5, 1)
dl = DeepLift(model)
model.zero_grad()
attributions, delta = dl.attribute(
input, baseline, target=target, return_convergence_delta=True
)
self._assert_attributions(model, attributions, input, baseline, delta, target)
# compare with integrated gradients
ig = IntegratedGradients(model)
attributions_ig = ig.attribute(input, baseline, target=target)
assertAttributionComparision(self, (attributions,), (attributions_ig,))
def test_softmax_classification_batch_multi_target(self) -> None:
num_in = 40
inputs = torch.arange(0.0, num_in * 3.0,
requires_grad=True).reshape(3, num_in)
baselines = torch.arange(1.0, num_in + 1).reshape(1, num_in)
model = SoftmaxDeepLiftModel(num_in, 20, 10)
dl = DeepLift(model)
self.softmax_classification(model, dl, inputs, baselines,
torch.tensor([2, 2, 2]))
def test_relu_linear_deeplift(self) -> None:
model = ReLULinearModel(inplace=True)
x1 = torch.tensor([[-10.0, 1.0, -5.0]], requires_grad=True)
x2 = torch.tensor([[3.0, 3.0, 1.0]], requires_grad=True)
inputs = (x1, x2)
baselines = (0, 0.0001)
# expected = [[[0.0, 0.0]], [[6.0, 2.0]]]
self._deeplift_assert(model, DeepLift(model), inputs, baselines)
def test_relu_deeplift(self):
x1 = torch.tensor([1.0], requires_grad=True)
x2 = torch.tensor([2.0], requires_grad=True)
b1 = torch.tensor([0.0], requires_grad=True)
b2 = torch.tensor([0.0], requires_grad=True)
inputs = (x1, x2)
baselines = (b1, b2)
model = ReLUDeepLiftModel()
self._deeplift_helper(model, DeepLift(model), inputs, baselines)
def test_relu_deeplift_batch(self) -> None:
x1 = torch.tensor([[1.0], [1.0], [1.0], [1.0]], requires_grad=True)
x2 = torch.tensor([[2.0], [2.0], [2.0], [2.0]], requires_grad=True)
b1 = torch.tensor([[0.0], [0.0], [0.0], [0.0]], requires_grad=True)
b2 = torch.tensor([[0.0], [0.0], [0.0], [0.0]], requires_grad=True)
inputs = (x1, x2)
baselines = (b1, b2)
model = ReLUDeepLiftModel()
self._deeplift_assert(model, DeepLift(model), inputs, baselines)
def test_tanh_deeplift(self) -> None:
x1 = torch.tensor([-1.0], requires_grad=True)
x2 = torch.tensor([-2.0], requires_grad=True)
b1 = torch.tensor([0.0], requires_grad=True)
b2 = torch.tensor([0.0], requires_grad=True)
inputs = (x1, x2)
baselines = (b1, b2)
model = TanhDeepLiftModel()
self._deeplift_assert(model, DeepLift(model), inputs, baselines)
def test_relu_linear_deeplift(self):
model = ReLULinearDeepLiftModel()
x1 = torch.tensor([[-10.0, 1.0, -5.0]], requires_grad=True)
x2 = torch.tensor([[3.0, 3.0, 1.0]], requires_grad=True)
b1 = torch.tensor([[0.0, 0.0, 0.0]], requires_grad=True)
b2 = torch.tensor([[0.0, 0.0, 0.0]], requires_grad=True)
inputs = (x1, x2)
baselines = (b1, b2)
# expected = [[[0.0, 0.0]], [[6.0, 2.0]]]
self._deeplift_helper(model, DeepLift(model), inputs, baselines)
def __init__(
self,
model: Module,
layer: Module,
multiply_by_inputs: bool = True,
) -> None:
r"""
Args:
model (nn.Module): The reference to PyTorch model instance. Model cannot
contain any in-place nonlinear submodules; these are not
supported by the register_full_backward_hook PyTorch API
starting from PyTorch v1.9.
layer (torch.nn.Module): Layer for which attributions are computed.
The size and dimensionality of the attributions
corresponds to the size and dimensionality of the layer's
input or output depending on whether we attribute to the
inputs or outputs of the layer.
multiply_by_inputs (bool, optional): Indicates whether to factor
model inputs' multiplier in the final attribution scores.
In the literature this is also known as local vs global
attribution. If inputs' multiplier isn't factored in
then that type of attribution method is also called local
attribution. If it is, then that type of attribution
method is called global.
More detailed can be found here:
https://arxiv.org/abs/1711.06104
In case of Layer DeepLift, if `multiply_by_inputs`
is set to True, final sensitivity scores
are being multiplied by
layer activations for inputs - layer activations for baselines.
This flag applies only if `custom_attribution_func` is
set to None.
"""
LayerAttribution.__init__(self, model, layer)
DeepLift.__init__(self, model)
self.model = model
self._multiply_by_inputs = multiply_by_inputs
def attribute(
self,
inputs: TensorOrTupleOfTensorsGeneric,
neuron_selector: Union[int, Tuple[Union[int, slice], ...], Callable],
baselines: BaselineType = None,
additional_forward_args: Any = None,
attribute_to_neuron_input: bool = False,
custom_attribution_func: Union[None, Callable[..., Tuple[Tensor, ...]]] = None,
) -> TensorOrTupleOfTensorsGeneric:
r"""
Args:
inputs (tensor or tuple of tensors): Input for which layer
attributions are computed. If forward_func takes a
single tensor as input, a single input tensor should be
provided. If forward_func takes multiple tensors as input,
a tuple of the input tensors should be provided. It is
assumed that for all given input tensors, dimension 0
corresponds to the number of examples (aka batch size),
and if multiple input tensors are provided, the examples
must be aligned appropriately.
neuron_selector (int, callable, or tuple of ints or slices):
Selector for neuron
in given layer for which attribution is desired.
Neuron selector can be provided as:
- a single integer, if the layer output is 2D. This integer
selects the appropriate neuron column in the layer input
or output
- a tuple of integers or slice objects. Length of this
tuple must be one less than the number of dimensions
in the input / output of the given layer (since
dimension 0 corresponds to number of examples).
The elements of the tuple can be either integers or
slice objects (slice object allows indexing a
range of neurons rather individual ones).
If any of the tuple elements is a slice object, the
indexed output tensor is used for attribution. Note
that specifying a slice of a tensor would amount to
computing the attribution of the sum of the specified
neurons, and not the individual neurons independantly.
- a callable, which should
take the target layer as input (single tensor or tuple
if multiple tensors are in layer) and return a neuron or
aggregate of the layer's neurons for attribution.
For example, this function could return the
sum of the neurons in the layer or sum of neurons with
activations in a particular range. It is expected that
this function returns either a tensor with one element
or a 1D tensor with length equal to batch_size (one scalar
per input example)
baselines (scalar, tensor, tuple of scalars or tensors, optional):
Baselines define reference samples that are compared with
the inputs. In order to assign attribution scores DeepLift
computes the differences between the inputs/outputs and
corresponding references.
Baselines can be provided as:
- a single tensor, if inputs is a single tensor, with
exactly the same dimensions as inputs or the first
dimension is one and the remaining dimensions match
with inputs.
- a single scalar, if inputs is a single tensor, which will
be broadcasted for each input value in input tensor.
- a tuple of tensors or scalars, the baseline corresponding
to each tensor in the inputs' tuple can be:
- either a tensor with matching dimensions to
corresponding tensor in the inputs' tuple
or the first dimension is one and the remaining
dimensions match with the corresponding
input tensor.
- or a scalar, corresponding to a tensor in the
inputs' tuple. This scalar value is broadcasted
for corresponding input tensor.
In the cases when `baselines` is not provided, we internally
use zero scalar corresponding to each input tensor.
Default: None
additional_forward_args (any, optional): If the forward function
requires additional arguments other than the inputs for
which attributions should not be computed, this argument
can be provided. It must be either a single additional
argument of a Tensor or arbitrary (non-tuple) type or a tuple
containing multiple additional arguments including tensors
or any arbitrary python types. These arguments are provided
to forward_func in order, following the arguments in inputs.
Note that attributions are not computed with respect
to these arguments.
Default: None
attribute_to_neuron_input (bool, optional): Indicates whether to
compute the attributions with respect to the neuron input
or output. If `attribute_to_neuron_input` is set to True
then the attributions will be computed with respect to
neuron's inputs, otherwise it will be computed with respect
to neuron's outputs.
Note that currently it is assumed that either the input
or the output of internal neuron, depending on whether we
attribute to the input or output, is a single tensor.
Support for multiple tensors will be added later.
Default: False
custom_attribution_func (callable, optional): A custom function for
computing final attribution scores. This function can take
at least one and at most three arguments with the
following signature:
- custom_attribution_func(multipliers)
- custom_attribution_func(multipliers, inputs)
- custom_attribution_func(multipliers, inputs, baselines)
In case this function is not provided, we use the default
logic defined as: multipliers * (inputs - baselines)
It is assumed that all input arguments, `multipliers`,
`inputs` and `baselines` are provided in tuples of same
length. `custom_attribution_func` returns a tuple of
attribution tensors that have the same length as the
`inputs`.
Default: None
Returns:
**attributions** or 2-element tuple of **attributions**, **delta**:
- **attributions** (*tensor* or tuple of *tensors*):
Computes attributions using Deeplift's rescale rule for
particular neuron with respect to each input feature.
Attributions will always be the same size as the provided
inputs, with each value providing the attribution of the
corresponding input index.
If a single tensor is provided as inputs, a single tensor is
returned. If a tuple is provided for inputs, a tuple of
corresponding sized tensors is returned.
Examples::
>>> # ImageClassifier takes a single input tensor of images Nx3x32x32,
>>> # and returns an Nx10 tensor of class probabilities.
>>> net = ImageClassifier()
>>> # creates an instance of LayerDeepLift to interpret target
>>> # class 1 with respect to conv4 layer.
>>> dl = NeuronDeepLift(net, net.conv4)
>>> input = torch.randn(1, 3, 32, 32, requires_grad=True)
>>> # Computes deeplift attribution scores for conv4 layer and neuron
>>> # index (4,1,2).
>>> attribution = dl.attribute(input, (4,1,2))
"""
dl = DeepLift(cast(Module, self.forward_func), self.multiplies_by_inputs)
dl.gradient_func = construct_neuron_grad_fn(
self.layer,
neuron_selector,
attribute_to_neuron_input=attribute_to_neuron_input,
)
# NOTE: using __wrapped__ to not log
return dl.attribute.__wrapped__( # type: ignore
dl, # self
inputs,
baselines,
additional_forward_args=additional_forward_args,
custom_attribution_func=custom_attribution_func,
)
def test_reusable_modules(self) -> None:
model = BasicModelWithReusableModules()
input = torch.rand(1, 3)
dl = DeepLift(model)
with self.assertRaises(RuntimeError):
dl.attribute(input, target=0)
