simple_cycle
ReservoirComputing.simple_cycle — Function
simple_cycle([rng], [T], dims...;
cycle_weight=0.1, return_sparse=false,
radius=nothing, kwargs...)Create a simple cycle reservoir (Rodan and Tino, 2011).
\[W_{i,j} = \begin{cases} r, & \text{if } i = j + 1,\;\; j \in [1, D_{\mathrm{res}} - 1], \\[4pt] r, & \text{if } i = 1,\;\; j = D_{\mathrm{res}}, \\[6pt] 0, & \text{otherwise.} \end{cases}\]
Arguments
rng: Random number generator. Default isUtils.default_rng()from WeightInitializers.T: Type of the elements in the reservoir matrix. Default isFloat32.dims: Dimensions of the reservoir matrix.
Keyword arguments
cycle_weight: Weight of the connections in the reservoir matrix. This can be provided as a single value or an array. In case it is provided as an array please make sure that the length of the array matches the length of the cycle you want to populate. Default is 0.1.radius: The desired spectral radius of the reservoir. Ifnothingis passed, no scaling takes place. Defaults tonothing.return_sparse: flag for returning asparsematrix.truerequiresSparseArraysto be loaded. Default isfalse.signs: Controls sign flips. UseRandomSigns,RegularSigns, orIrrationalDigitSigns. Passnothingto leave signs unchanged. Default isnothing.
Examples
Default call:
julia> res_matrix = simple_cycle(5, 5)
5×5 Matrix{Float32}:
0.0 0.0 0.0 0.0 0.1
0.1 0.0 0.0 0.0 0.0
0.0 0.1 0.0 0.0 0.0
0.0 0.0 0.1 0.0 0.0
0.0 0.0 0.0 0.1 0.0Changing weights:
julia> res_matrix = simple_cycle(5, 5; cycle_weight=0.99)
5×5 Matrix{Float32}:
0.0 0.0 0.0 0.0 0.99
0.99 0.0 0.0 0.0 0.0
0.0 0.99 0.0 0.0 0.0
0.0 0.0 0.99 0.0 0.0
0.0 0.0 0.0 0.99 0.0Changing weights to a custom array:
julia> cycle_weights = Float32[0.2, 0.4, 0.6, 0.8, 1.0];
julia> res_matrix = simple_cycle(5, 5; cycle_weight = cycle_weights);
julia> res_matrix[2, 1] == 0.2f0 && res_matrix[1, 5] == 1.0f0
trueChanging sign of the weights with different sign patterns:
julia> irrational_matrix = simple_cycle(5, 5; signs = IrrationalDigitSigns());
julia> bernoulli_matrix = simple_cycle(MersenneTwister(123), 5, 5; signs = RandomSigns());
julia> all(abs.(irrational_matrix[irrational_matrix .!= 0]) .== 0.1f0) && all(abs.(bernoulli_matrix[bernoulli_matrix .!= 0]) .== 0.1f0)
trueReturning as sparse:
julia> using SparseArrays
julia> res_matrix = simple_cycle(5, 5; return_sparse=true)
5×5 SparseMatrixCSC{Float32, Int64} with 5 stored entries:
⋅ ⋅ ⋅ ⋅ 0.1
0.1 ⋅ ⋅ ⋅ ⋅
⋅ 0.1 ⋅ ⋅ ⋅
⋅ ⋅ 0.1 ⋅ ⋅
⋅ ⋅ ⋅ 0.1 ⋅References
- Rodan, A. and Tino, P. (2011). Minimum Complexity Echo State Network. IEEE Transactions on Neural Networks 22, 131–144.