dynamicslab / normalformae

Normal form Autoencoder

Constructing low-dimensional parameterized representations of high dimensional dynamics using normal forms as a building block by using an autoencoder framework. Neural network training is implemented with Flux.jl, a Julia library. Paper available on arXiv

Reproduce results

Download datasets from here, and extract contents to NormalFormAE/NFAEdata. Use the scripts in run to reproduce the results from the paper.

Note you need CUDA to run this package.

How to use

• If you have Linux/Mac, run the following on your terminal to install Julia in one command

bash -ci "$(curl -fsSL https://raw.githubusercontent.com/abelsiqueira/jill/master/jill.sh)"

from here.

• Clone this package and enter the directory. Run julia on your terminal.
• Now run the following:

 julia> ] activate .
 julia> ] instantiate

which will automatically install the necessary Julia packages you need.

• Run an example (tests coming soon) via the terminal or REPL Shell mode. Note to run in the REPL Shell mode, you need to use the backspace/delete key to exit out of Pkg mode, and then type a ;. Find out more about running Julia files in the Julia docs.

julia -i run/run_nf.jl

normalformae's Issues

Not running. BSON file corrupted.

There is problem with the BSON files that makes it problematic to run.

Dependecies make it impossible to use

When following the instructions one runs into the problem that the precompilation always fails. Probably because there is some issue with compatibilities with a newer version of julia (in my case v1.7.3)
A workaround was to delete the manifest.toml, run ] instantiate and then ] build

Not sure how to avoid it from the start