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¶
- Citation — how to cite ResDAG and the methods it implements
- Contributing — development setup