Reduce¶

class continual.Reduce(reduce='sum')[source]¶

Reduce multiple input streams to a single using the selected function

For instance, here is how it is used to sum streams in a residual connection:

residual = co.Sequential(
    co.Broadcast(2),
    co.Parallel(
        co.Conv3d(32, 32, kernel_size=3, padding=1),
        co.Delay(2),
    ),
    co.Reduce("sum"),
)

A user-defined can be passed as well:

from functools import reduce

def my_sum(inputs):
    return reduce(torch.Tensor.add, inputs[1:], inputs[0])

residual = co.Sequential(
    co.Broadcast(2),
    co.Parallel(
        co.Conv3d(32, 32, kernel_size=3, padding=1),
        co.Delay(2),
    ),
    co.Reduce(my_sum),
)

Parameters:: reduce (Union[str, Callable[[Sequence[Tensor]], Tensor]]) – Reduce function. Either one of [“sum”, “channel”, “mul”, “max”] or user-defined function mapping a sequence of tensors to a single one.