output_size

hybrid_learning.concepts.models.model_extension.output_size(model, input_size, has_batch_dim=False, layer_id=None)[source]

Feed dummy input of input_size to model to determine the output size of the layer with ID layer_id. If no layer ID is given, the size of the model final output is returned.

Will raise if the model output is not a single tensor. This essentially is a wrapper around output_sizes() that uses the fact that the model also occurs in its torch.nn.Module.named_modules() listing, with ID ''. So, the model output can also be obtained by registering a hook to ''.

Parameters

  • model (Module) – the model the output of which is to be investigated

  • input_size (Sequence[int]) – a single, all-zero tensor of that size is fed to the model

  • layer_id (Optional[str]) – the ID of the layer to inspect; if not given, all layers (i.e. '') are inspected

  • has_batch_dim (bool) – whether the given input_size already features a batch dimension; added if not

Return type

Size