Giter VIP home page Giter VIP logo

vgdf's Introduction

Cross-domain Policy Adaptation via Value-guided Data Filtering

This is the official code for the paper "Cross-domain Policy Adaptation via Value-guided Data Filtering".

Train

Train VGDF in HalfCheetah - broken back thigh with:

python script/train_vgdf.py --env halfcheetah --seeds 12

Train VGDF_BC in HalfCheetah - no thighs with:

python script/train_vgdf_bc.py --env halfcheetah_morph --seeds 12

You can tune any hyperparameters in the config for experiments.

References

@inproceedings{xu2023cross,
  title={Cross-domain Policy Adaptation via Value-guided Data Filtering},
  author={Xu, Kang and Bai, Chenjia and Ma, Xiaoteng and Wang, Dong and Zhao, Bin and Wang, Zhen and Li, Xuelong and Li, Wei},
  booktitle = {Advances in Neural Information Processing Systems},
  year = {2023}
}

vgdf's People

Contributors

kangxu023 avatar kavka1 avatar

Stargazers

avatar avatar avatar avatar

Watchers

avatar

vgdf's Issues

Target likelihood goes to zeros

Hi @kangxu023, @Kavka1,

Thank you for your amazing work.
I'm trying your code and run the experiments for the vgdf_bc with 4 environments: halfcheetah - broken back thigh, halfcheetah - no thighs, hopper - broken hips and hopper - short feet. Unfortunately, I only got the successful results for the hopper - short feet environment. For all 3 other environments, it seems like the agent learn nothing. I have checked the logging and saw that the target likelihood tends to go to zeros. Beside that, the critic loss comes to very large value just after 100s training steps. Here are the logging results
image
image
image
image

Could you please help me to solve this issue?
Thank you so much.

Environment setup

Hi @kangxu023 ,

Thank you for your awesome works and paper.
I am running your code and wonder about the python environment setup (dependencies, libs, etc.) cause I couldn't find any setup files or instructions about that.
Could you please help me to install the necessary libraries to run your code and experiments.

Thank you a lot and have a great day.

Recommend Projects

  • React photo React

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

  • Vue.js photo Vue.js

    🖖 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. 📊📈🎉

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 ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.