Giter VIP home page Giter VIP logo

run-skeleton-run's Introduction

Run-Skeleton-Run

Reason8.ai PyTorch solution for 3rd place NIPS RL 2017 challenge.

Theano version

Additional thanks to Mikhail Pavlov for collaboration.

Agent policies

no-flip-state-action

alt text

flip-state-action

alt text

How to setup environment?

  1. sh setup_conda.sh
  2. source activate opensim-rl

Would like to test baselines? (Need MPI support)

  1. sudo apt-get install openmpi-bin openmpi-doc libopenmpi-dev 3+. sh setup_env_mpi.sh

OR like DDPG agents? 3. sh setup_env.sh

  1. Congrats! Now you are ready to check our agents.

Run DDPG agent

CUDA_VISIBLE_DEVICES="" PYTHONPATH=. python ddpg/train.py \
    --logdir ./logs_ddpg \
    --num-threads 4 \
    --ddpg-wrapper \
    --skip-frames 5 \
    --fail-reward -0.2 \
    --reward-scale 10 \
    --flip-state-action \
    --actor-layers 64-64 --actor-layer-norm --actor-parameters-noise \
    --actor-lr 0.001 --actor-lr-end 0.00001 \
    --critic-layers 64-32 --critic-layer-norm \
    --critic-lr 0.002 --critic-lr-end 0.00001 \
    --initial-epsilon 0.5 --final-epsilon 0.001 \
    --tau 0.0001

Evaluate DDPG agent

CUDA_VISIBLE_DEVICES="" PYTHONPATH=./ python ddpg/submit.py \
    --restore-actor-from ./logs_ddpg/actor_state_dict.pkl \
    --restore-critic-from ./logs_ddpg/critic_state_dict.pkl \
    --restore-args-from ./logs_ddpg/args.json \
    --num-episodes 10

Run TRPO/PPO agent

CUDA_VISIBLE_DEVICES="" PYTHONPATH=. python ddpg/train.py \
    --agent ppo \
    --logdir ./logs_baseline \
    --baseline-wrapper \
    --skip-frames 5 \
    --fail-reward -0.2 \
    --reward-scale 10

Citation

Please cite the following paper if you feel this repository useful.

@article{run_skeleton,
  title={Run, skeleton, run: skeletal model in a physics-based simulation},
  author = {Mikhail Pavlov, Sergey Kolesnikov and Sergey M.~Plis},
  journal={AAAI Spring Symposium Series},
  year={2018}
}

run-skeleton-run's People

Contributors

scitator avatar shmuma avatar wassname avatar

Stargazers

 avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar

Watchers

 avatar  avatar  avatar  avatar  avatar  avatar

run-skeleton-run's Issues

Parameter Space Noise or NoisyNet?

I am confused about the noise you used in the code. In your writeup at arxiv, you said you used parameter space noise(https://arxiv.org/abs/1706.01905). However, I think you actually used NoisyNet in your code. In common/modules/NoisyLinear.py, you also said you used the NoisyNet. So could you tell where you used the parameter space noise? Am I misunderstanding? I am really confused. Need your help!

Package versions

It's great to have battle tested RL implementation available. But without knowing the exact package versions you used people might get differen't results. For example other projects have found pytorch versions can cause models to diverge.

So could you please run pip freeze and conda list in the environment that you trained the ddpg model, and share the outputs? (We could ideally turn it into a requirements.txt file or put it in your setup_*.sh scripts).

Thanks for sharing this!

[Question?]ModuleNotFoundError: No module named 'common.logger'

I've got an error as follows. I think that the reason is logger.py is not in common directory. Could the authors fix it?

File "Run-Skeleton-Run/ddpg/model.py", line 12, in
from common.logger import Logger
ModuleNotFoundError: No module named 'common.logger'

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.