l5kit.planning.vectorized.global_graph module

class l5kit.planning.vectorized.global_graph.MLP(input_dim: int, hidden_dim: int, output_dim: int, num_layers: int)

Bases: torch.nn.modules.module.Module

Very simple multi-layer perceptron (also called FFN)

forward(x: torch.Tensor) → torch.Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

reset_parameters() → None

training: bool

class l5kit.planning.vectorized.global_graph.MultiheadAttentionGlobalHead(d_model: int, num_timesteps: int, num_outputs: int, nhead: int = 8, dropout: float = 0.1)

Bases: torch.nn.modules.module.Module

Global graph making use of multi-head attention.

forward(inputs: torch.Tensor, type_embedding: torch.Tensor, mask: torch.Tensor) → Tuple[torch.Tensor, Optional[torch.Tensor]]

Model forward:

Parameters

• inputs – model inputs

• type_embedding – type embedding describing the different input types

• mask – availability mask

:return tuple of outputs, attention

training: bool

class l5kit.planning.vectorized.global_graph.VectorizedEmbedding(embedding_dim: int)

Bases: torch.nn.modules.module.Module

forward(data_batch: Dict[str, torch.Tensor]) → torch.Tensor

Model forward: embed the given elements based on their type.

Assumptions: - agent of interest is the first one in the batch - other agents follow - then we have polylines (lanes)

training: bool