ConcatAggregationCombination

class ConcatAggregationCombination(aggregation: str | Callable[[Tensor], Tensor] | None = None, dim: int = -1)[source]

Combine representation by concatenation followed by an aggregation along the same axis.

Initialize the combination.

Parameters:

aggregation (str | Callable[[Tensor], Tensor] | None) – the aggregation, or a hint thereof, cf. class_resolver.contrib.torch.aggregation_resolver
dim (int) – the concatenation and reduction dimension.

Methods Summary

`forward`(xs)	Combine a sequence of individual representations.
`iter_extra_repr`()	Iterate over the components of the `extra_repr()`.

Methods Documentation

Combine a sequence of individual representations.

Parameters:: xs (Sequence[Tensor]) – shape: (*batch_dims, *input_dims_i) the individual representations
Returns:: shape: (*batch_dims, *output_dims) a combined representation
Return type:: Tensor

iter_extra_repr() → Iterable[str][source]

Iterate over the components of the extra_repr().

This method is typically overridden. A common pattern would be

def iter_extra_repr(self) -> Iterable[str]:
    yield from super().iter_extra_repr()
    yield "<key1>=<value1>"
    yield "<key2>=<value2>"