Embed to Control implementation in PyTorch

e2c-pytorch
Paper can be found here: https://arxiv.org/abs/1506.07365
You will need a patched version of OpenAI Gym in order to generate the dataset. See https://github.com/ethanluoyc/gym/tree/pendulum_internal
For the planar task, we use code from. The source code of the repository has been modified for our needs and included under e2c/e2c_tf
.
What’s included ?
- E2C model, VAE and AE baselines. Allow configuration for different network architecture for the different setups (see Appendix of the paper).
TODO
- Documentation, tests… (Soon to follow)
GitHub