Giter Club home page Giter Club logo

goms's Introduction

实验计划:

state: 10: 3 position, 3 linear velocity, 4 angular velocity
action: 3 position control
goal: 3 position


实验1:    训练:
          输入 state + next_state
          输出 action
          测试:
          输入 state + goal
          输出 action
结果1:    对于 fetch_reacher 任务来说效果不错

实验2:    训练:
          输入 state + goal
          输出 action
          测试:
          输入 state + goal
          输出 action
结果2:    对于 fetch_pick_and_place 任务来说效果不错

实验3:    在实验2上减少训练的样本数量


问题1:怎么让 random 采集到的数据 make sense ?!!
答案1:手动编写控制代码

问题2:训练得到的 policy 的成功率为0
答案2:让原来的回归任务改成分类任务
      训练4个分类器,其中3个是3分类任务,一个是2分类任务

      上面的答案不对!!
      正确的原因是提供的特征的维度太多,造成了维度爆炸。
      解决办法是把原来的长度为25的observation缩短到长度为6

      结果:
      无论是分类任务还是回归任务,准确率都可以接近100%

问题3:如何从gym中得到图片
答案3:
        1. 设置render的mode参数
            env.render(mode='rgb_array')
        2. 去除多余的显示信息
            ~/Documents/mujoco-py/mujoco_py/mjviewer.py中
           class MjViewer 中
           self._hide_overlay = True
        3. 更改图片的大小
            ~/anaconda3/envs/tf-cpu/lib/python3.6/site-packages/gym/envs/mujoco/mujoco_env.py中
           class:MujocoEnv function:render中
           width, height = 1744, 992



注意注意!!!!
    之前的代码 “forward-consistent-feature-reduced-GSP.py” 有错误
    在 train 函数的 next_state_feed 的获取中,我应该把 j+36 而不是加 32 !!!!


问题4:如果模型A包含模型B,如何在模型A中加载模型B预训练好的权重?

    我的一个担忧是:如果在模型A中直接调用模型B,(即把模型B当做一个层来调用)那么在模型A中加载模型B的权重会成功吗?
    更保险地做法是:在模型A中加入和模型B一模一样的层,而不是直接调用B。



问题5:  用两个LSTM来实现 auto-encoder 存在的问题

    decoder可能什么也没学到

    有两个解决办法:
    1,decoder不用LSTM
    2,对状态序列进行降采样


gym 里面的一个bug是:
    when using render(), the firs time the claw is very low
    but if not using render(), the problem is gone


之前犯了一个非常愚蠢地错误是在我的第一个版本的top-sub policy 中,我加载权重的时候设置了 by_name = True, 导致权重加载不正确

goms's People

Contributors

diodiodada avatar

Watchers

James Cloos avatar  avatar

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.