def test_basic_relu_multi_input(self) -> None:
model = BasicModel2()
input1 = torch.tensor([[3.0]])
input2 = torch.tensor([[1.0]], requires_grad=True)
baseline1 = torch.tensor([[0.0]])
baseline2 = torch.tensor([[0.0]])
inputs = (input1, input2)
baselines = (baseline1, baseline2)
gs = GradientShap(model)
n_samples = 30000
attributions, delta = cast(
Tuple[Tuple[Tensor, ...], Tensor],
gs.attribute(
inputs,
baselines=baselines,
n_samples=n_samples,
return_convergence_delta=True,
),
)
_assert_attribution_delta(self, inputs, attributions, n_samples, delta)
ig = IntegratedGradients(model)
attributions_ig = ig.attribute(inputs, baselines=baselines)
self._assert_shap_ig_comparision(attributions, attributions_ig)
def test_basic_scalar_baselines_multi_input(self):
inputs = (torch.ones(10, 3), torch.rand(10, 4))
baselines_scalars = (0.0, 0.0)
baselines_tensors = (
torch.zeros(1, 3),
torch.zeros(10, 4),
)
model = BasicLinearModel()
model.eval()
gradient_shap = GradientShap(model)
attributions_base_scalars, _ = gradient_shap.attribute(
inputs,
baselines_scalars,
n_samples=50,
return_convergence_delta=True)
attributions_base_tnsrs, _ = gradient_shap.attribute(
inputs,
baselines_tensors,
n_samples=50,
return_convergence_delta=True)
assertTensorAlmostEqual(self, attributions_base_scalars[0],
attributions_base_tnsrs[0])
assertTensorAlmostEqual(self, attributions_base_scalars[1],
attributions_base_tnsrs[1])
def test_classification_baselines_as_function(self) -> None:
num_in = 40
inputs = torch.arange(0.0, num_in * 2.0).reshape(2, num_in)
def generate_baselines() -> Tensor:
return torch.arange(0.0, num_in * 4.0).reshape(4, num_in)
def generate_baselines_with_inputs(inputs: Tensor) -> Tensor:
inp_shape = cast(Tuple[int, ...], inputs.shape)
return torch.arange(
0.0,
inp_shape[1] * 2.0,
).reshape(2, inp_shape[1])
def generate_baselines_returns_array() -> ndarray:
return np.arange(0.0, num_in * 4.0).reshape(4, num_in)
# 10-class classification model
model = SoftmaxModel(num_in, 20, 10)
model.eval()
model.zero_grad()
gradient_shap = GradientShap(model)
n_samples = 10
attributions, delta = gradient_shap.attribute(
inputs,
baselines=generate_baselines,
target=torch.tensor(1),
n_samples=n_samples,
stdevs=0.009,
return_convergence_delta=True,
)
_assert_attribution_delta(self, (inputs, ), (attributions, ),
n_samples, delta)
attributions, delta = gradient_shap.attribute(
inputs,
baselines=generate_baselines_with_inputs,
target=torch.tensor(1),
n_samples=n_samples,
stdevs=0.00001,
return_convergence_delta=True,
)
_assert_attribution_delta(self, (inputs, ), (attributions, ),
n_samples, delta)
with self.assertRaises(AssertionError):
attributions, delta = gradient_shap.attribute(
inputs,
baselines=generate_baselines_returns_array,
target=torch.tensor(1),
n_samples=n_samples,
stdevs=0.00001,
return_convergence_delta=True,
)
def test_basic_multi_input(self):
batch_size = 10
x1 = torch.ones(batch_size, 3)
x2 = torch.ones(batch_size, 4)
inputs = (x1, x2)
batch_size_baselines = 20
baselines = (
torch.zeros(batch_size_baselines, 3),
torch.zeros(batch_size_baselines, 4),
)
class Net(nn.Module):
def __init__(self):
super().__init__()
self.linear = nn.Linear(7, 1)
def forward(self, x1, x2):
return self.linear(torch.cat((x1, x2), dim=-1))
model = Net()
model.eval()
model.zero_grad()
np.random.seed(0)
torch.manual_seed(0)
gradient_shap = GradientShap(model)
n_samples = 50
attributions, delta = gradient_shap.attribute(
(x1, x2),
baselines,
n_samples=n_samples,
return_convergence_delta=True)
attributions_without_delta = gradient_shap.attribute((x1, x2),
baselines)
self._assert_attribution_delta(inputs, attributions, n_samples, delta)
# Compare with integrated gradients
ig = IntegratedGradients(model)
baselines = (torch.zeros(batch_size, 3), torch.zeros(batch_size, 4))
attributions_ig = ig.attribute(inputs, baselines=baselines)
self._assert_shap_ig_comparision(attributions, attributions_ig)
# compare attributions retrieved with and without
# `return_convergence_delta` flag
for attribution, attribution_without_delta in zip(
attributions, attributions_without_delta):
assertTensorAlmostEqual(self, attribution,
attribution_without_delta)
def test_basic_multi_input(self) -> None:
batch_size = 10
x1 = torch.ones(batch_size, 3)
x2 = torch.ones(batch_size, 4)
inputs = (x1, x2)
batch_size_baselines = 20
baselines = (
torch.zeros(batch_size_baselines, 3),
torch.zeros(batch_size_baselines, 4),
)
model = BasicLinearModel()
model.eval()
model.zero_grad()
np.random.seed(0)
torch.manual_seed(0)
gradient_shap = GradientShap(model)
n_samples = 50
attributions, delta = cast(
Tuple[Tuple[Tensor, ...], Tensor],
gradient_shap.attribute(inputs,
baselines,
n_samples=n_samples,
return_convergence_delta=True),
)
attributions_without_delta = gradient_shap.attribute((x1, x2),
baselines)
_assert_attribution_delta(self, inputs, attributions, n_samples, delta)
# Compare with integrated gradients
ig = IntegratedGradients(model)
baselines = (torch.zeros(batch_size, 3), torch.zeros(batch_size, 4))
attributions_ig = ig.attribute(inputs, baselines=baselines)
self._assert_shap_ig_comparision(attributions, attributions_ig)
# compare attributions retrieved with and without
# `return_convergence_delta` flag
for attribution, attribution_without_delta in zip(
attributions, attributions_without_delta):
assertTensorAlmostEqual(self, attribution,
attribution_without_delta)
def test_basic_multilayer_compare_w_inp_features(self):
model = BasicModel_MultiLayer()
model.eval()
inputs = torch.tensor([[10.0, 20.0, 10.0]])
baselines = torch.randn(30, 3)
gs = GradientShap(model)
expected, delta = gs.attribute(
inputs, baselines, target=0, return_convergence_delta=True
)
self.setUp()
self._assert_attributions(
model,
model.linear0,
inputs,
baselines,
0,
expected,
expected_delta=delta,
attribute_to_layer_input=True,
)
def test_basic_multi_input_wo_mutliplying_by_inputs(self) -> None:
batch_size = 10
x1 = torch.ones(batch_size, 3)
x2 = torch.ones(batch_size, 4)
inputs = (x1, x2)
batch_size_baselines = 20
baselines = (
torch.ones(batch_size_baselines, 3) + 2e-5,
torch.ones(batch_size_baselines, 4) + 2e-5,
)
model = BasicLinearModel()
model.eval()
model.zero_grad()
np.random.seed(0)
torch.manual_seed(0)
gradient_shap = GradientShap(model)
gradient_shap_wo_mutliplying_by_inputs = GradientShap(
model, multiply_by_inputs=False)
n_samples = 50
attributions = cast(
Tuple[Tuple[Tensor, ...], Tensor],
gradient_shap.attribute(
inputs,
baselines,
n_samples=n_samples,
stdevs=0.0,
),
)
attributions_wo_mutliplying_by_inputs = cast(
Tuple[Tuple[Tensor, ...], Tensor],
gradient_shap_wo_mutliplying_by_inputs.attribute(
inputs,
baselines,
n_samples=n_samples,
stdevs=0.0,
),
)
assertTensorAlmostEqual(
self,
attributions_wo_mutliplying_by_inputs[0] *
(x1 - baselines[0][0:1]),
attributions[0],
)
assertTensorAlmostEqual(
self,
attributions_wo_mutliplying_by_inputs[1] *
(x2 - baselines[1][0:1]),
attributions[1],
)
def test_classification(self):
num_in = 40
inputs = torch.arange(0.0, num_in * 2.0).reshape(2, num_in)
baselines = torch.arange(0.0, num_in * 4.0).reshape(4, num_in)
target = torch.tensor(1)
# 10-class classification model
model = SoftmaxModel(num_in, 20, 10)
model.eval()
model.zero_grad()
gradient_shap = GradientShap(model)
n_samples = 10
attributions, delta = gradient_shap.attribute(
inputs,
baselines=baselines,
target=target,
n_samples=n_samples,
stdevs=0.009,
return_convergence_delta=True,
)
_assert_attribution_delta(self, (inputs, ), (attributions, ),
n_samples, delta)
# try to call `compute_convergence_delta` externally
with self.assertRaises(AssertionError):
gradient_shap.compute_convergence_delta(attributions,
inputs,
baselines,
target=target)
# now, let's expand target and choose random baselines from `baselines` tensor
rand_indices = np.random.choice(baselines.shape[0],
inputs.shape[0]).tolist()
chosen_baselines = baselines[rand_indices]
target_extendes = torch.tensor([1, 1])
external_delta = gradient_shap.compute_convergence_delta(
attributions, chosen_baselines, inputs, target=target_extendes)
_assert_delta(self, external_delta)
# Compare with integrated gradients
ig = IntegratedGradients(model)
baselines = torch.arange(0.0, num_in * 2.0).reshape(2, num_in)
attributions_ig = ig.attribute(inputs,
baselines=baselines,
target=target)
self._assert_shap_ig_comparision((attributions, ), (attributions_ig, ))
def attribute(
self,
inputs: TensorOrTupleOfTensorsGeneric,
neuron_selector: Union[int, Tuple[Union[int, slice], ...], Callable],
baselines: Union[TensorOrTupleOfTensorsGeneric,
Callable[..., TensorOrTupleOfTensorsGeneric]],
n_samples: int = 5,
stdevs: float = 0.0,
additional_forward_args: Any = None,
attribute_to_neuron_input: bool = False,
) -> TensorOrTupleOfTensorsGeneric:
r"""
Args:
inputs (tensor or tuple of tensors): Input for which SHAP attribution
values 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, 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 (tensor, tuple of tensors, callable):
Baselines define the starting point from which expectation
is computed and can be provided as:
- a single tensor, if inputs is a single tensor, with
the first dimension equal to the number of examples
in the baselines' distribution. The remaining dimensions
must match with input tensor's dimension starting from
the second dimension.
- a tuple of tensors, if inputs is a tuple of tensors,
with the first dimension of any tensor inside the tuple
equal to the number of examples in the baseline's
distribution. The remaining dimensions must match
the dimensions of the corresponding input tensor
starting from the second dimension.
- callable function, optionally takes `inputs` as an
argument and either returns a single tensor
or a tuple of those.
It is recommended that the number of samples in the baselines'
tensors is larger than one.
n_samples (int, optional): The number of randomly generated examples
per sample in the input batch. Random examples are
generated by adding gaussian random noise to each sample.
Default: `5` if `n_samples` is not provided.
stdevs (float, or a tuple of floats optional): The standard deviation
of gaussian noise with zero mean that is added to each
input in the batch. If `stdevs` is a single float value
then that same value is used for all inputs. If it is
a tuple, then it must have the same length as the inputs
tuple. In this case, each stdev value in the stdevs tuple
corresponds to the input with the same index in the inputs
tuple.
Default: 0.0
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 can contain a tuple of ND tensors or
any arbitrary python type of any shape.
In case of the ND tensor the first dimension of the
tensor must correspond to the batch size. It will be
repeated for each `n_steps` for each randomly generated
input sample.
Note that the gradients 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
Returns:
**attributions** or 2-element tuple of **attributions**, **delta**:
- **attributions** (*tensor* or tuple of *tensors*):
Attribution score computed based on GradientSHAP 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()
>>> neuron_grad_shap = NeuronGradientShap(net, net.linear2)
>>> input = torch.randn(3, 3, 32, 32, requires_grad=True)
>>> # choosing baselines randomly
>>> baselines = torch.randn(20, 3, 32, 32)
>>> # Computes gradient SHAP of first neuron in linear2 layer
>>> # with respect to the input's of the network.
>>> # Attribution size matches input size: 3x3x32x32
>>> attribution = neuron_grad_shap.attribute(input, neuron_ind=0
baselines)
"""
gs = GradientShap(self.forward_func, self.multiplies_by_inputs)
gs.gradient_func = construct_neuron_grad_fn(
self.layer,
neuron_selector,
self.device_ids,
attribute_to_neuron_input=attribute_to_neuron_input,
)
# NOTE: using __wrapped__ to not log
return gs.attribute.__wrapped__( # type: ignore
gs, # self
inputs,
baselines,
n_samples=n_samples,
stdevs=stdevs,
additional_forward_args=additional_forward_args,
)