Metadata-Version: 2.1
Name: jax-fixedpoint
Version: 0.0.4
Summary: Implicit and competitive differentiation in JAX.
Home-page: UNKNOWN
Author: Clement Gehring
Author-email: fax-dev@gehring.io
License: MIT License
Platform: UNKNOWN
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: MacOS :: MacOS X
Classifier: Operating System :: Microsoft :: Windows
Classifier: Operating System :: POSIX
Classifier: Programming Language :: C++
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.5
Description-Content-Type: text/markdown
Requires-Dist: numpy
Requires-Dist: scipy
Requires-Dist: absl-py
Requires-Dist: jax
Requires-Dist: jaxlib
Requires-Dist: hypothesis

# fax: fixed-point jax 

Implicit and [competitive differentiation](https://optrl2019.github.io/assets/accepted_papers/70.pdf) in [JAX](https://github.com/google/jax).

Our "competitive differentiation" approach uses [Competitive Gradient Descent](https://arxiv.org/abs/1905.12103) to solve the equality-constrained nonlinear program associated with the fixed-point problem. A standalone implementation of CGD is provided under [fax/competitive/cga.py](fax/competitive/cga.py) and the equality-constrained solver derived from it can be accessed via `fax.constrained.cga_lagrange_min` or `fax.constrained.cga_ecp`. An implementation of implicit differentiation based on [Christianson's](https://doi.org/10.1080/10556789408805572) two-phases reverse accumulation algorithm can also be obtained with the function `fax.implicit.two_phase_solver`.

See [fax/constrained/constrained_test.py](fax/constrained/constrained_test.py) for examples. Please note that the API is subject to change.

## References

Citing competitive differentiation:

```
@inproceedings{bacon2019optrl,
  author={Pierre-Luc Bacon, Florian Schaefer, Clement Gehring, Animashree Anandkumar, Emma Brunskill},
  title={A Lagrangian Method for Inverse Problems in Reinforcement Learning},
  booktitle={NeurIPS Optimization Foundations for Reinforcement Learning Workshop},
  year={2019},
  url={http://lis.csail.mit.edu/pubs/bacon-optrl-2019.pdf},
  keywords={Optimization, Reinforcement Learning, Lagrangian}
}
```

Citing this repo:

```
@misc{gehring2019fax,
  author = {Clement Gehring, Pierre-Luc Bacon, Florian Schaefer},
  title = {{FAX: differentiating fixed point problems in JAX}},
  note = {Available at: https://github.com/gehring/fax},
  year = {2019}
}
```

