alibi.explainers.backends.pytorch.cfrl_base module

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

class alibi.explainers.backends.pytorch.cfrl_base.PtCounterfactualRLDataset(X, preprocessor, predictor, conditional_func, batch_size)

Bases: CounterfactualRLDataset, Dataset

Pytorch backend datasets.

__init__(X, preprocessor, predictor, conditional_func, batch_size)

Constructor.

Parameters:

• X (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.

alibi.explainers.backends.pytorch.cfrl_base.add_noise(Z_cf, noise, act_low, act_high, step, exploration_steps, device, **kwargs)

Add noise to the counterfactual embedding.

Parameters:

• Z_cf (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 (device) – Device to send data to.

Return type:

Tensor

Returns:

Z_cf_tilde – Noised counterfactual embedding.

alibi.explainers.backends.pytorch.cfrl_base.consistency_loss(Z_cf_pred, Z_cf_tgt)

Default 0 consistency loss.

Parameters:

• Z_cf_pred (Tensor) – Counterfactual embedding prediction.

• Z_cf_tgt (Tensor) – Counterfactual embedding target.

Returns:

0 consistency loss.

alibi.explainers.backends.pytorch.cfrl_base.data_generator(X, encoder_preprocessor, predictor, conditional_func, batch_size, shuffle, num_workers, **kwargs)

Constructs a tensorflow data generator.

Parameters:

• X (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.

alibi.explainers.backends.pytorch.cfrl_base.decode(Z, decoder, device, **kwargs)

Decodes an embedding tensor.

Parameters:

• Z (Tensor) – Embedding tensor to be decoded.

• decoder (Module) – Pretrained decoder network.

• device (device) – Device to sent data to.

Returns:

Embedding tensor decoding.

alibi.explainers.backends.pytorch.cfrl_base.encode(X, encoder, device, **kwargs)

Encodes the input tensor.

Parameters:

• X (Tensor) – Input to be encoded.

• encoder (Module) – Pretrained encoder network.

• device (device) – Device to send data to.

Returns:

Input encoding.

alibi.explainers.backends.pytorch.cfrl_base.generate_cf(Z, Y_m, Y_t, C, encoder, decoder, actor, device, **kwargs)

Generates counterfactual embedding.

Parameters:

• Z (Tensor) – Input embedding tensor.

• Y_m (Tensor) – Input classification label.

• Y_t (Tensor) – Target counterfactual classification label.

• C (Optional[Tensor]) – Conditional tensor.

• encoder (Module) – Pretrained encoder network.

• decoder (Module) – Pretrained decoder network.

• actor (Module) – Actor network. The model generates the counterfactual embedding.

• device (device) – Device object to be used.

Return type:

Tensor

Returns:

Z_cf – Counterfactual embedding.

alibi.explainers.backends.pytorch.cfrl_base.get_actor(hidden_dim, output_dim)

Constructs the actor network.

Parameters:

• hidden_dim (int) – Actor’s hidden dimension

• output_dim (int) – Actor’s output dimension.

Return type:

Module

Returns:

Actor network.

alibi.explainers.backends.pytorch.cfrl_base.get_critic(hidden_dim)

Constructs the critic network.

Parameters:

hidden_dim (int) – Critic’s hidden dimension.

Return type:

Module

Returns:

Critic network.

alibi.explainers.backends.pytorch.cfrl_base.get_device()

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

Return type:

device

Returns:

Device to be used.

alibi.explainers.backends.pytorch.cfrl_base.get_optimizer(model, lr=0.001)

Constructs default Adam optimizer.

Return type:

Optimizer

Returns:

Default optimizer.

alibi.explainers.backends.pytorch.cfrl_base.load_model(path)

Loads a model and its optimizer.

Parameters:

path (Union[str, PathLike]) – Path to the loading location.

Return type:

Module

Returns:

Loaded model.

alibi.explainers.backends.pytorch.cfrl_base.save_model(path, model)

Saves a model and its optimizer.

Parameters:

• path (Union[str, PathLike]) – Path to the saving location.

• model (Module) – Model to be saved.

Return type:

None

alibi.explainers.backends.pytorch.cfrl_base.set_seed(seed=13)

Sets a seed to ensure reproducibility.

Parameters:

seed (int) – Seed to be set.

alibi.explainers.backends.pytorch.cfrl_base.sparsity_loss(X_hat_cf, X)

Default L1 sparsity loss.

Parameters:

• X_hat_cf (Tensor) – Auto-encoder counterfactual reconstruction.

• X (Tensor) – Input instance

Return type:

Dict[str, Tensor]

Returns:

L1 sparsity loss.

alibi.explainers.backends.pytorch.cfrl_base.to_numpy(X)

Converts given tensor to numpy array.

Parameters:

X (Union[List, ndarray, Tensor, None]) – Input tensor to be converted to numpy array.

Return type:

Union[List, ndarray, None]

Returns:

Numpy representation of the input tensor.

alibi.explainers.backends.pytorch.cfrl_base.to_tensor(X, device, **kwargs)

Converts tensor to torch.Tensor

Return type:

Optional[Tensor]

Returns:

torch.Tensor conversion.

alibi.explainers.backends.pytorch.cfrl_base.update_actor_critic(encoder, decoder, critic, actor, optimizer_critic, optimizer_actor, sparsity_loss, consistency_loss, coeff_sparsity, coeff_consistency, X, X_cf, Z, Z_cf_tilde, Y_m, Y_t, C, R_tilde, device, **kwargs)

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

Parameters:

• encoder (Module) – Pretrained encoder network.

• decoder (Module) – Pretrained decoder network.

• critic (Module) – Critic network.

• actor (Module) – Actor network.

• optimizer_critic (Optimizer) – Critic’s optimizer.

• optimizer_actor (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 (ndarray) – Input array.

• X_cf (ndarray) – Counterfactual array.

• Z (ndarray) – Input embedding.

• Z_cf_tilde (ndarray) – Noised counterfactual embedding.

• Y_m (ndarray) – Input classification label.

• Y_t (ndarray) – Target counterfactual classification label.

• C (Optional[ndarray]) – Conditional tensor.

• R_tilde (ndarray) – Noised counterfactual reward.

• device (device) – Torch device object.

• **kwargs – Other arguments. Not used.

Returns:

Dictionary of losses.