chirho_diffeqpy

An experimental diffeqpy (DifferentialEquations.jl) backend for chirho's dynamical systems module.

See docs/source/performance_comparison.ipynb for a comparison between chirho's default TorchDiffEq backend and the DiffEqPy backend provided here. That notebook also covers usage examples and differences between DiffEqPy and TorchDiffEq compatible definitions of dynamics.

See this guide on selecting a solver algorithm for your problem. Algorithms and tolerances can be specified at solver instantiation time, as shown below. All arguments are optional, but must be specified as key word arguments.

from diffeqpy import de
from chirho_diffeqpy import DiffEqPy
solver_instance = DiffEqPy(reltol=1e-6, abstol=1e-8, alg=de.Tsit5())

Installation

pip install git+https://github.com/BasisResearch/chirho_diffeqpy.git

Note that on the first import of the package, a large number of julia packages will be installed and precompiled in a shared environment called "diffeqpy". This can take some time (up to 30 minutes).

Installing with Tests

The module comes with some internal tests, but primarily relies on a reparametrization of chirho's test suite to assess correctness. In order to run these tests, we currently require that chirho is manually installed as a local source distribution before running the tests. To run tests locally:

1. create a virtual environment
2. clone this repo: git clone https://github.com/BasisResearch/chirho_diffeqpy.git
3. install from source: pip install -e './chirho_diffeqpy[test]'
4. clone chirho: git clone https://github.com/BasisResearch/chirho.git; cd chirho; git reset --hard 0f5dae6; cd ..
5. install chirho from source: pip install -e './chirho[dynamical,test]'

Finally, tests can be run like so: pytest chirho_diffeqpy/tests