Skip to content

Build · Topology (graph)

spectral_cascade

Generates a graph of N disconnected components (cliques) of sizes 1..N, where n = N(N+1)/2. Each clique is fully connected (or complete in the directed sense), with deterministic weights in {+1, -1}: - off-diagonal: (-1)^(i + j) - diagonal (if self_loops=True): (-1)^i

spectral_cascade

Connectivity and weight matrix on a small reservoir.
from resdag.layers import ESNLayer

reservoir = ESNLayer(500, feedback_size=3, topology="spectral_cascade", spectral_radius=0.9)
parameter default
spectral_radius —
self_loops False

Override any parameter with the tuple spec — topology=("spectral_cascade", {...}) — or pre-configure with get_topology("spectral_cascade", ...). Scaling to a target spectral radius happens after construction, whatever the source.

See also