nfan(n_out, n_in=1) -> Tuple
nfan(dims...)
nfan(dims::Tuple)

For a layer characterized by dimensions dims, return a tuple (fan_in, fan_out), where fan_in is the number of input neurons connected to an output one, and fan_out is the number of output neurons connected to an input one.

This function is mainly used by weight initializers, e.g., [ kaiming_normal]( Flux.kaiming_normal).

Examples

			julia> layer = Dense(10, 20);

julia> Flux.nfan(size(layer.weight))
(10, 20)

julia> layer = Conv((3, 3), 2=>10);

julia> Flux.nfan(size(layer.weight))
(18, 90)