Skip to content

Latest commit

 

History

History
35 lines (27 loc) · 1.07 KB

README.md

File metadata and controls

35 lines (27 loc) · 1.07 KB
Raw

SLIP: Learning to predict in unknown dynamical systems with long-term memory

This repository includes an implementation of the spectral LDS improper predictor (SLIP) from our paper

Paria Rashidinejad, Jiantao Jiao, and Stuart Russell. "SLIP: Learning to Predict in Unknown Dynamical Systems with Long-Term Memory." NeurIPS 2020 (oral).

Prerequisites

  • Python (>= 3.5)
  • NumPy (>= 1.6)
  • SciPy (>= 0.17.0)
  • Matplotlib (>= 1.5.1)
  • Scikit-Learn

How to run the code

To run a simulation, in the LDS settings cell specify

  • the LDS parameters or let the function generate_random_LDS sample them randomly based on dimensions;
  • the time horizon;
  • the number of iterations;
  • the number of filters (based on horizon) and regularization parameter.
python SLIP.py

Cite

@article{rashidinejad2020slip,
  title={SLIP: Learning to Predict in Unknown Dynamical Systems with Long-Term Memory},
  author={Rashidinejad, Paria and Jiao, Jiantao and Russell, Stuart},
  journal={arXiv preprint arXiv:2010.05899},
  year={2020}
}