The goal is infer parameters in a model that can predict future output, given new input. We are testing the following models:
- Nonlinear AutoRegressive model with eXogenous input (NARX; see ForneyLab node code)
- Nonlinear AutoRegressive Moving Average model with eXogenous input (NARMAX; see ForneyLab node code)
These models are standard in the control systems community. Typically, (recursive) least-squares or some other form of frequentist estimation is used. In this repo, we employ Free Energy Minimisation.
We run a series of verification experiments and a validation experiment on the Silverbox data set from the Nonlinear Benchmark.
Questions, comments and general feedback can be directed to the issues tracker.