This is the code repository for the paper A trajectory's guide to the state space - learning missing terms in bifurcating dynamical systems (Vortmeyer-Kley, Nieters, Pipa, 2021).

Code organization

The core components of the experiments are organized in the local package GeneralizedDyanmicsFromData inluding a locally defined environment that can be activated and installed using the Julia Package Manager.

For each of the dynamical systems used in the paper, and example file is provided to run one UDE with our standard neural network model and a SInDy identification of the missing term approximated by the neural network.

For each dynamical system, we calculated the distribution of losses after training from 100 random initializations of the neural network. The corresponding files for each system and each loss are provided.

The implementations of the loss functions can be found in src/losslib.jl, the descriptions of the UDEs and ODEs can be found in src/eqlib.jl.