* ODENS:

This is the authors' implementation of Optimism Driven Exploration for
Nonlinear Systems [1], provided here to ensure that results are
reproducible and that all parameters omitted from the article are
available. For now, the code is not appropriately packaged for general
re-use, has not been thoroughly cleaned, and does not include
documentation.

* Installing:

  1. unzip odens.zip
  2. install mosek
  3. install nvida cuda drivers
  4. install current version of scikits.cuda
  Alternatively:
  4. install our fork of scikits.cuda from
     https://github.com/teodor-moldovan/scikits.cuda

* Usage:

  1. make dynamicstest
  2. make learning

* Notes:

- At the time of writing the official scikits.cuda package did not
support all the needed functionality provided by the nvidia cuda
drives. We have forked the package to include the missing features.
- Inference for the Dirichlet process mixture model and parts of the
dynamics models have been implemented for GPU (cuda) parallel
computation via pycuda and scikits.cuda. GPU kernels are dynamically
generated, compiled and cached on disk for re-use. This is a one-time
operation that is typically slow.
- Hard-coded parameters corresponding to specific GPU hardware
capabilities will be sub-optimal for other hardware models. For this
reason, do not expect to replicate the running times presented in the
article.

* References:

[1] Teodor Mihai Moldovan, Sergey Levine, Michael I. Jordan, Pieter
Abbeel, "Optimism-Driven Exploration for Nonlinear Systems",
Proceedings of the IEEE International Conference on Robotics and
Automation (ICRA), 2015