alibi_detect.utils.keops.kernels

DeepKernel

Inherits from: Module

Constructor

DeepKernel(self, proj: torch.nn.modules.module.Module, kernel_a: Union[torch.nn.modules.module.Module, typing_extensions.Literal['rbf']] = 'rbf', kernel_b: Union[torch.nn.modules.module.Module, typing_extensions.Literal['rbf'], NoneType] = 'rbf', eps: Union[float, typing_extensions.Literal['trainable']] = 'trainable') -> None

Name	Type	Default	Description
proj	torch.nn.modules.module.Module		The projection to be applied to the inputs before applying kernel_a
kernel_a	Union[torch.nn.modules.module.Module, Literal[rbf]]	'rbf'	The kernel to apply to the projected inputs. Defaults to a Gaussian RBF with trainable bandwidth.
kernel_b	Union[torch.nn.modules.module.Module, Literal[rbf], None]	'rbf'	The kernel to apply to the raw inputs. Defaults to a Gaussian RBF with trainable bandwidth. Set to None in order to use only the deep component (i.e. eps=0).
eps	Union[float, Literal[trainable]]	'trainable'	The proportion (in [0,1]) of weight to assign to the kernel applied to raw inputs. This can be either specified or set to 'trainable'. Only relavent if kernel_b is not None.

Properties

Property	Type	Description
eps	torch.Tensor

Methods

forward

forward(x_proj: pykeops.torch.lazytensor.LazyTensor.LazyTensor, y_proj: pykeops.torch.lazytensor.LazyTensor.LazyTensor, x: Optional[pykeops.torch.lazytensor.LazyTensor.LazyTensor] = None, y: Optional[pykeops.torch.lazytensor.LazyTensor.LazyTensor] = None) -> pykeops.torch.lazytensor.LazyTensor.LazyTensor

Name	Type	Default	Description
x_proj	pykeops.torch.lazytensor.LazyTensor.LazyTensor
y_proj	pykeops.torch.lazytensor.LazyTensor.LazyTensor
x	Optional[pykeops.torch.lazytensor.LazyTensor.LazyTensor]	None
y	Optional[pykeops.torch.lazytensor.LazyTensor.LazyTensor]	None

Returns

• Type: pykeops.torch.lazytensor.LazyTensor.LazyTensor

from_config

from_config(config)

Name	Type	Default	Description
config

get_config

get_config() -> dict

Returns

• Type: dict

GaussianRBF

Inherits from: Module

Constructor

GaussianRBF(self, sigma: Optional[torch.Tensor] = None, init_sigma_fn: Optional[Callable] = None, trainable: bool = False) -> None

Name	Type	Default	Description
sigma	Optional[torch.Tensor]	None	Bandwidth used for the kernel. Needn't be specified if being inferred or trained. Can pass multiple values to eval kernel with and then average.
init_sigma_fn	Optional[Callable]	None	Function used to compute the bandwidth sigma. Used when sigma is to be inferred. The function's signature should match :py:func:~alibi_detect.utils.keops.kernels.sigma_mean, meaning that it should take in the lazy tensors x, y and dist and return a tensor sigma.
trainable	bool	False	Whether or not to track gradients w.r.t. sigma to allow it to be trained.

Properties

Property	Type	Description
sigma	torch.Tensor

Methods

forward

forward(x: pykeops.torch.lazytensor.LazyTensor.LazyTensor, y: pykeops.torch.lazytensor.LazyTensor.LazyTensor, infer_sigma: bool = False) -> pykeops.torch.lazytensor.LazyTensor.LazyTensor

Name	Type	Default	Description
x	pykeops.torch.lazytensor.LazyTensor.LazyTensor
y	pykeops.torch.lazytensor.LazyTensor.LazyTensor
infer_sigma	bool	False

Returns

• Type: pykeops.torch.lazytensor.LazyTensor.LazyTensor

from_config

from_config(config)

Instantiates a kernel from a config dictionary.

Name	Type	Default	Description
config			A kernel config dictionary.

get_config

get_config() -> dict

Returns a serializable config dict (excluding the input_sigma_fn, which is serialized in alibi_detect.saving).

Returns

• Type: dict

Functions

sigma_mean

sigma_mean(x: pykeops.torch.lazytensor.LazyTensor.LazyTensor, y: pykeops.torch.lazytensor.LazyTensor.LazyTensor, dist: pykeops.torch.lazytensor.LazyTensor.LazyTensor, n_min: int = 100) -> torch.Tensor

Set bandwidth to the mean distance between instances x and y.

Name	Type	Default	Description
x	pykeops.torch.lazytensor.LazyTensor.LazyTensor		LazyTensor of instances with dimension [Nx, 1, features] or [batch_size, Nx, 1, features]. The singleton dimension is necessary for broadcasting.
y	pykeops.torch.lazytensor.LazyTensor.LazyTensor		LazyTensor of instances with dimension [1, Ny, features] or [batch_size, 1, Ny, features]. The singleton dimension is necessary for broadcasting.
dist	pykeops.torch.lazytensor.LazyTensor.LazyTensor		LazyTensor with dimensions [Nx, Ny] or [batch_size, Nx, Ny] containing the pairwise distances between x and y.
n_min	int	100	In order to check whether x equals y after squeezing the singleton dimensions, we check if the diagonal of the distance matrix (which is a lazy tensor from which the diagonal cannot be directly extracted) consists of all zeros. We do this by computing the k-min distances and k-argmin indices over the columns of the distance matrix. We then check if the distances on the diagonal of the distance matrix are all zero or not. If they are all zero, then we do not use these distances (zeros) when computing the mean pairwise distance as bandwidth. If Nx becomes very large, it is advised to set n_min to a low enough value to avoid OOM issues. By default we set it to 100 instances.

Returns

• Type: torch.Tensor