alibi.explainers.backends.pytorch.cfrl_base

This module contains utility functions for the Counterfactual with Reinforcement Learning base class, :py:class:alibi.explainers.cfrl_base for the Pytorch backend.

Constants

TYPE_CHECKING

TYPE_CHECKING: bool = False

bool(x) -> bool

Returns True when the argument x is true, False otherwise. The builtins True and False are the only two instances of the class bool. The class bool is a subclass of the class int, and cannot be subclassed.

PtCounterfactualRLDataset

Inherits from: CounterfactualRLDataset, ABC, Dataset, Generic

Pytorch backend datasets.

Constructor

PtCounterfactualRLDataset(self, X: numpy.ndarray, preprocessor: Callable, predictor: Callable, conditional_func: Callable, batch_size: int) -> None

Name	Type	Default	Description
X	numpy.ndarray		Array of input instances. The input should NOT be preprocessed as it will be preprocessed when calling the preprocessor function.
preprocessor	Callable		Preprocessor function. This function correspond to the preprocessing steps applied to the auto-encoder model.
predictor	Callable		Prediction function. The classifier function should expect the input in the original format and preprocess it internally in the predictor if necessary.
conditional_func	Callable		Conditional function generator. Given an preprocessed input array, the functions generates a conditional array.
batch_size	int		Dimension of the batch used during training. The same batch size is used to infer the classification labels of the input dataset.

Functions

add_noise

add_noise(Z_cf: torch.Tensor, noise: NormalActionNoise, act_low: float, act_high: float, step: int, exploration_steps: int, device: torch.device, kwargs) -> torch.Tensor

Add noise to the counterfactual embedding.

Name	Type	Default	Description
Z_cf	torch.Tensor		Counterfactual embedding.
noise	NormalActionNoise		Noise generator object.
act_low	float		Action lower bound.
act_high	float		Action upper bound.
step	int		Training step.
exploration_steps	int		Number of exploration steps. For the first exploration_steps, the noised counterfactual embedding is sampled uniformly at random.
device	torch.device		Device to send data to.

Returns

• Type: torch.Tensor

consistency_loss

consistency_loss(Z_cf_pred: torch.Tensor, Z_cf_tgt: torch.Tensor)

Default 0 consistency loss.

Name	Type	Default	Description
Z_cf_pred	torch.Tensor		Counterfactual embedding prediction.
Z_cf_tgt	torch.Tensor		Counterfactual embedding target.

data_generator

data_generator(X: numpy.ndarray, encoder_preprocessor: Callable, predictor: Callable, conditional_func: Callable, batch_size: int, shuffle: bool, num_workers: int, kwargs)

Constructs a tensorflow data generator.

Name	Type	Default	Description
X	numpy.ndarray		Array of input instances. The input should NOT be preprocessed as it will be preprocessed when calling the preprocessor function.
encoder_preprocessor	Callable		Preprocessor function. This function correspond to the preprocessing steps applied to the encoder/auto-encoder model.
predictor	Callable		Prediction function. The classifier function should expect the input in the original format and preprocess it internally in the predictor if necessary.
conditional_func	Callable		Conditional function generator. Given an preprocessed input array, the functions generates a conditional array.
batch_size	int		Dimension of the batch used during training. The same batch size is used to infer the classification labels of the input dataset.
shuffle	bool		Whether to shuffle the dataset each epoch. True by default.
num_workers	int		Number of worker processes to be created.
**kwargs			Other arguments. Not used.

decode

decode(Z: torch.Tensor, decoder: torch.nn.modules.module.Module, device: torch.device, kwargs)

Decodes an embedding tensor.

Name	Type	Default	Description
Z	torch.Tensor		Embedding tensor to be decoded.
decoder	torch.nn.modules.module.Module		Pretrained decoder network.
device	torch.device		Device to sent data to.

encode

encode(X: torch.Tensor, encoder: torch.nn.modules.module.Module, device: torch.device, kwargs)

Encodes the input tensor.

Name	Type	Default	Description
X	torch.Tensor		Input to be encoded.
encoder	torch.nn.modules.module.Module		Pretrained encoder network.
device	torch.device		Device to send data to.

generate_cf

generate_cf(Z: torch.Tensor, Y_m: torch.Tensor, Y_t: torch.Tensor, C: Optional[torch.Tensor], encoder: torch.nn.modules.module.Module, decoder: torch.nn.modules.module.Module, actor: torch.nn.modules.module.Module, device: torch.device, kwargs) -> torch.Tensor

Generates counterfactual embedding.

Name	Type	Default	Description
Z	torch.Tensor		Input embedding tensor.
Y_m	torch.Tensor		Input classification label.
Y_t	torch.Tensor		Target counterfactual classification label.
C	Optional[torch.Tensor]		Conditional tensor.
encoder	torch.nn.modules.module.Module		Pretrained encoder network.
decoder	torch.nn.modules.module.Module		Pretrained decoder network.
actor	torch.nn.modules.module.Module		Actor network. The model generates the counterfactual embedding.
device	torch.device		Device object to be used.

Returns

• Type: torch.Tensor

get_actor

get_actor(hidden_dim: int, output_dim: int) -> torch.nn.modules.module.Module

Constructs the actor network.

Name	Type	Default	Description
hidden_dim	int		Actor's hidden dimension
output_dim	int		Actor's output dimension.

Returns

• Type: torch.nn.modules.module.Module

get_critic

get_critic(hidden_dim: int) -> torch.nn.modules.module.Module

Constructs the critic network.

Name	Type	Default	Description
hidden_dim	int		Critic's hidden dimension.

Returns

• Type: torch.nn.modules.module.Module

get_device

get_device() -> torch.device

Checks if cuda is available. If available, use cuda by default, else use cpu.

Returns

• Type: torch.device

get_optimizer

get_optimizer(model: torch.nn.modules.module.Module, lr: float = 0.001) -> torch.optim.optimizer.Optimizer

Constructs default Adam optimizer.

Name	Type	Default	Description
model	torch.nn.modules.module.Module
lr	float	0.001

Returns

• Type: torch.optim.optimizer.Optimizer

load_model

load_model(path: Union[str, os.PathLike]) -> torch.nn.modules.module.Module

Loads a model and its optimizer.

Name	Type	Default	Description
path	Union[str, os.PathLike]		Path to the loading location.

Returns

• Type: torch.nn.modules.module.Module

save_model

save_model(path: Union[str, os.PathLike], model: torch.nn.modules.module.Module) -> None

Saves a model and its optimizer.

Name	Type	Default	Description
path	Union[str, os.PathLike]		Path to the saving location.
model	torch.nn.modules.module.Module		Model to be saved.

Returns

• Type: None

set_seed

set_seed(seed: int = 13)

Sets a seed to ensure reproducibility.

Name	Type	Default	Description
seed	int	13	Seed to be set.

sparsity_loss

sparsity_loss(X_hat_cf: torch.Tensor, X: torch.Tensor) -> Dict[str, torch.Tensor]

Default L1 sparsity loss.

Name	Type	Default	Description
X_hat_cf	torch.Tensor		Auto-encoder counterfactual reconstruction.
X	torch.Tensor		Input instance

Returns

• Type: Dict[str, torch.Tensor]

to_numpy

to_numpy(X: Union[List[Any], numpy.ndarray, torch.Tensor, None]) -> Union[List[Any], numpy.ndarray, None]

Converts given tensor to numpy array.

Name	Type	Default	Description
X	Union[List[Any], numpy.ndarray, torch.Tensor, None]		Input tensor to be converted to numpy array.

Returns

• Type: Union[List[Any], numpy.ndarray, None]

to_tensor

to_tensor(X: Union[numpy.ndarray, torch.Tensor], device: torch.device, kwargs) -> Optional[torch.Tensor]

Converts tensor to torch.Tensor

Name	Type	Default	Description
X	Union[numpy.ndarray, torch.Tensor]
device	torch.device

Returns

• Type: Optional[torch.Tensor]

update_actor_critic

update_actor_critic(encoder: torch.nn.modules.module.Module, decoder: torch.nn.modules.module.Module, critic: torch.nn.modules.module.Module, actor: torch.nn.modules.module.Module, optimizer_critic: torch.optim.optimizer.Optimizer, optimizer_actor: torch.optim.optimizer.Optimizer, sparsity_loss: Callable, consistency_loss: Callable, coeff_sparsity: float, coeff_consistency: float, X: numpy.ndarray, X_cf: numpy.ndarray, Z: numpy.ndarray, Z_cf_tilde: numpy.ndarray, Y_m: numpy.ndarray, Y_t: numpy.ndarray, C: Optional[numpy.ndarray], R_tilde: numpy.ndarray, device: torch.device, kwargs)

Training step. Updates actor and critic networks including additional losses.

Name	Type	Default	Description
encoder	torch.nn.modules.module.Module		Pretrained encoder network.
decoder	torch.nn.modules.module.Module		Pretrained decoder network.
critic	torch.nn.modules.module.Module		Critic network.
actor	torch.nn.modules.module.Module		Actor network.
optimizer_critic	torch.optim.optimizer.Optimizer		Critic's optimizer.
optimizer_actor	torch.optim.optimizer.Optimizer		Actor's optimizer.
sparsity_loss	Callable		Sparsity loss function.
consistency_loss	Callable		Consistency loss function.
coeff_sparsity	float		Sparsity loss coefficient.
coeff_consistency	float		Consistency loss coefficient
X	numpy.ndarray		Input array.
X_cf	numpy.ndarray		Counterfactual array.
Z	numpy.ndarray		Input embedding.
Z_cf_tilde	numpy.ndarray		Noised counterfactual embedding.
Y_m	numpy.ndarray		Input classification label.
Y_t	numpy.ndarray		Target counterfactual classification label.
C	Optional[numpy.ndarray]		Conditional tensor.
R_tilde	numpy.ndarray		Noised counterfactual reward.
device	torch.device		Torch device object.
**kwargs			Other arguments. Not used.