alibi_detect.models.pytorch package

class alibi_detect.models.pytorch.TransformerEmbedding(model_name_or_path, embedding_type, layers=None)

Bases: Module

forward(tokens)

Return type:

Tensor

alibi_detect.models.pytorch.trainer(model, loss_fn, dataloader, device, optimizer=torch.optim.Adam, learning_rate=0.001, preprocess_fn=None, epochs=20, reg_loss_fn=<function <lambda>>, verbose=1)

Train PyTorch model.

Parameters:

• model (Union[Module, Sequential]) – Model to train.

• loss_fn (Callable) – Loss function used for training.

• dataloader (DataLoader) – PyTorch dataloader.

• device (device) – Device used for training.

• optimizer (Callable) – Optimizer used for training.

• learning_rate (float) – Optimizer’s learning rate.

• preprocess_fn (Optional[Callable]) – Preprocessing function applied to each training batch.

• epochs (int) – Number of training epochs.

• reg_loss_fn (Callable) – The regularisation term reg_loss_fn(model) is added to the loss function being optimized.

• verbose (int) – Whether to print training progress.

Return type:

None

Submodules

• alibi_detect.models.pytorch.embedding module
  • TransformerEmbedding
    • TransformerEmbedding.forward()
  • hidden_state_embedding()
• alibi_detect.models.pytorch.gmm module
  • GMMModel
    • GMMModel.__init__()
    • GMMModel.forward()
• alibi_detect.models.pytorch.trainer module
  • trainer()