/SMBAE

Model-Based Action Exploration for Learning Dynamic Motion Skills learning code.

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

Learn

This package contains all of the python code used for learning. The code is based on Lasagne which is based on Theano.

Dependancies

  1. installDependanciesPython3.sh

Install On Windows

  1. Install Anaconda
  2. Follow the setup instruction for Theano which are
	conda install numpy scipy mkl-service libpython m2w64-toolchain <nose> <nose-parameterized> <sphinx> <pydot-ng>

For GPU training

  1. sudo apt-get install nvidia-cuda-toolkit nvidia-cuda-dev nvidia-modprobe
    These libraries are needed to compile code for the GPU as well as to check what GPU devices are available

NOTE: Ran into this issue on Ubuntu 16.04 (Theano/Theano#4425) As a temporary workaround, I use the following hack:

Add cmd.append('-D_FORCE_INLINES') just before p = subprocess.Popen( in the file nvcc_compiler.py

Using The system

python3 trainModel.py --config=settings/particleSim/PPO/PPO.json

Running meta simulations

These simulations are designed to sample a few simulations in order to get a more reasonable average of the performance of a method.

python3 tuneHyperParameters.py --config=tests/settings/particleSim/PPO/PPO_KERAS_Tensorflow.json --metaConfig=settings/hyperParamTuning/elementAI.json --meta_sim_samples=5 --meta_sim_threads=5 --tuning_threads=2

References

  1. https://github.com/Newmu/Theano-Tutorials
  2. https://github.com/spragunr/deep_q_rl