Echo State Networks Initializers

This page lists all initializers available in ReservoirComputing.jl. Clicking on any initializer name will take you to its dedicated documentation page, where full details and examples are provided.

Input layers

• chebyshev_mapping: Creates an input matrix using sine initialization followed by Chebyshev iterative mapping.
• informed_init: Builds an informed ESN input layer allocating input vs. model channels based on γ-split.
• logistic_mapping: Generates an input layer using sine initialization followed by logistic-map recursion.
• minimal_init: Creates a uniform-weight input layer with signs determined by a sampling scheme.
• modified_lm: Builds an input-expanding logistic-map chain for each input dimension.
• scaled_rand: Produces a uniformly scaled random input matrix with per-column or global scaling.
• weighted_init: Creates a block-structured weighted input layer with random weights per block.
• weighted_minimal: Generates a deterministic block-structured weighted input layer with optional sign sampling.

Reservoirs

• block_diagonal: Constructs a block-diagonal reservoir with constant-valued square blocks.
• chaotic_init: Generates a reservoir from a digital chaotic adjacency graph with rescaled spectral radius.
• cycle_jumps: Builds a cycle reservoir augmented with periodic jump connections.
• delay_line: Creates a delay-line reservoir using fixed offsets from the diagonal.
• delayline_backward: Produces a delay-line reservoir with additional backward (feedback) connections.
• double_cycle: Creates two interlaced directed cycles (upper & lower) with independent weights.
• forward_connection: Builds a reservoir where each node connects forward by two steps.
• low_connectivity: Creates a low-degree random (or enforced cycle) connectivity reservoir.
• pseudo_svd: Builds a reservoir by iteratively perturbing a diagonal matrix using pseudo-SVD rotations.
• rand_sparse: Generates a random sparse reservoir with controlled sparsity and spectral radius.
• selfloop_cycle: Builds a simple cycle reservoir enhanced with self-loops on all nodes.
• selfloop_delayline_backward: Combines delay line, self-loops, and backward offsets into one architecture.
• selfloop_backward_cycle: Creates a cycle where odd nodes self-loop and even nodes have forward/backward links.
• selfloop_forwardconnection: Adds self-loops onto a forward-connection reservoir (stride-2).
• simple_cycle: Builds a basic directed ring reservoir with uniform weights.
• true_doublecycle: Constructs two overlapping cycles (forward + backward) using Rodan-style cycle rules.

Building functions

• add_jumps!: Inserts jump connections at fixed intervals into an existing reservoir.
• backward_connection!: Adds backward (feedback) connections at a fixed shift.
• delay_line!: Writes delay connections into an existing matrix at a specified diagonal offset.
• reverse_simple_cycle!: Adds a reversed directed cycle (descending indices) to a matrix.
• scale_radius!: Rescales a reservoir matrix to match a target spectral radius.
• self_loop!: Adds self-loop weights along the diagonal of an existing matrix.
• simple_cycle!: Writes a directed cycle pattern into a preallocated reservoir matrix.