Skip to content

Project

Ecosystem

pytorch_symbolic

ResDAG's composition layer is built on pytorch_symbolic by Szymon Mikler. It provides the symbolic tracing that turns layer calls on placeholder tensors into executable model graphs — the mechanism behind Input, the functional API, and the model class. Without it, the central premise of this library — reservoir models composed as arbitrary DAGs — would not exist in its current form. Credit where it is due.

TSDynamics

TSDynamics is a companion library of dynamical systems by the same author: a catalog of chaotic ODEs, maps, and delay systems with a uniform integration interface, plus Lyapunov analysis utilities.

The two projects are designed to pair: TSDynamics generates the systems, ResDAG forecasts them. The benchmark systems in this documentation — the landing-page animation, the forecast figures — are TSDynamics systems, and the integration will deepen as both libraries grow (shared benchmark suites, data loaders for its systems, and reservoirs for continuous-time dynamics).

# the intended pairing
# pip install tsdynamics resdag
data = ...        # trajectory from a TSDynamics system, (batch, time, dim)
model = ...       # a ResDAG model

See also