search

openai / vime

Code for the paper "Curiosity-driven Exploration in Deep Reinforcement Learning via Bayesian Neural Networks"
https://arxiv.org/abs/1605.09674
342 stars 90 forks source link

Could you add some information to recreate the state space figure? #3

Open mrdrozdov opened 7 years ago

mrdrozdov commented 7 years ago

This is the figure I'm referring to:

screenshot 2017-03-31 23 17 12
ghost commented 7 years ago

Yep, same as @mrdrozdov,

mrdrozdov commented 7 years ago

I guess the task is mountain car [1] and the state space is position velocity, so if you kept a running list of all positions/velocities, maybe could make something like this diagram.

[1] https://github.com/openai/gym/wiki/MountainCar-v0

ghost commented 7 years ago

Thanks @mrdrozdov, I'm just starting to work on this. Seems like this repo is not actively maintained?

Given the nature of the paper, I'm wondering whether it might be worthwhile using Edward to experiment with?

mrdrozdov commented 7 years ago

It's not clear to me that Edward would have a clear benefit. I think it's probably better to use this repo/tf as it's already implemented!

SJTUGuofei commented 6 years ago

Hi, @mrdrozdov and @AjayTalati

I want to get this example code run on my own machine, and my environment is ubuntu16.04 with anaconda2 installed, and I have already well configured the rllab and mujoco. But I still can get this run, so here I have a few questions:

  1. Should I start the rllab3 env. before running the vime code? If I first start the rllab3, then it always comes out "ImportError: No module named 'cPickle'". And if I just run the code in python2 env., It always comes out "ImportError: No module named mako.template". but I have been already with the mako.template installed.
  2. Could you please tell me your configuration procedure? Thank you so much!