minimal_init

minimal_init([rng], [T], dims...; weight = 0.1, signs = nothing)

Create a dense matrix with magnitudes determined by weight (Rodan and Tino, 2011). An optional sign pattern controls which weights have their signs flipped.

Arguments

• rng: Random number generator. Default is Utils.default_rng() from WeightInitializers.
• T: Type of the elements in the reservoir matrix. Default is Float32.
• dims: Dimensions of the matrix. Should follow res_size x in_size.

Keyword arguments

• weight: The weight used to fill the layer matrix. Default is 0.1.
• signs: Controls sign flips. Use RandomSigns, RegularSigns, or IrrationalDigitSigns. Pass nothing to leave signs unchanged. Default is nothing.

Examples

Standard call:

julia> using Random

julia> using ReservoirComputing: IrrationalDigitSigns, RandomSigns, minimal_init

julia> res_input = minimal_init(MersenneTwister(14), Float32, 8, 3);

julia> size(res_input) == (8, 3) && all(abs.(res_input) .== 0.1f0)
true

Applying a sign pattern from an irrational number:

julia> res_input = minimal_init(
           MersenneTwister(15), Float32, 8, 3; signs = IrrationalDigitSigns());

julia> size(res_input) == (8, 3) && all(abs.(res_input) .== 0.1f0)
true

Changing the probability of preserving each sign

julia> low_probability = minimal_init(
           MersenneTwister(16), Float32, 8, 3; signs = RandomSigns(0.1));

julia> high_probability = minimal_init(
           MersenneTwister(16), Float32, 8, 3; signs = RandomSigns(0.8));

julia> count(>(0), low_probability) < count(>(0), high_probability)
true