-
Notifications
You must be signed in to change notification settings - Fork 243
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
投稿: 哈密顿蒙特卡洛算法为什么会发散或停止 徐晓鹏 #897
base: master
Are you sure you want to change the base?
Conversation
pull request, equation did not display, original blog: https://statisticalcomputing.github.io/hmc1.html
感谢投稿!请将 PR 题目改为"投稿:标题+作者的形式",如: "投稿: 数据通灵术 杜亚磊"。不知锐哥是否有时间审稿?@zhanruicai |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
请作者先按照以上建议修改。如有问题随时沟通。
- 机器学习 | ||
--- | ||
|
||
本文讨论哈密顿蒙特卡洛算法的稳定性问题及其解决方法。 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
建议作者加入一段关于哈密顿算法的背景介绍。可以从一个实际的小例子讲起,为什么要用哈密顿算法。哈密顿算法是什么样的?有什么优点?以及本文需要解决的问题是什么。
毕竟是科普系列的文章,我们需要假设读者没有相关的背景知识。
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
比如说,相比于MCMC,哈密顿的特点是?
(MCMC的背景也须介绍)
|
||
本文讨论哈密顿蒙特卡洛算法的稳定性问题及其解决方法。 | ||
|
||
这里的主要思想是:一个哈密顿仿真的总能量守恒,则仿真是稳定的。 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
如何定义能量?稳定是什么意思?
1. 接受概率公式 | ||
2. 能量守恒 | ||
|
||
,我们逐一进行讨论。 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
这里的算法能否不用图片截图?看起来别扭。
|
||
在统计计算中,我们一般研究第二种情况。这意味着不同位置对应的动能之和不同,对应的动量分布也不同。因此在仿真算法中,如果在不同位置,以相同分布(比如标准正态)生成动量初始值,那么就破坏了能量守恒原则,也违反了统计(力学)的基本假设。 | ||
|
||
受到krauth的statis tical mechanics algorithm and computation的启发,我们采用扔皮球的方式进行讨论。 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Typos:
Author name: Werner Krauth.
Book name: Statistical Mechanics: Algorithms and Computations.
|
||
这里的主要思想是:一个哈密顿仿真的总能量守恒,则仿真是稳定的。 | ||
|
||
比如如下这个仿真,红色的是位置-势能轨迹,黑色是位置轨迹,而蓝色是势能(对应概率)的三倍标准差界限。 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
什么是势能?什么又是势能对应的概率?
|
||
反之,如果总能量不守恒: | ||
|
||
1. 总能量持续变大,则最大势能也会持续变大,导致仿真发散; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same issues. I dont understand what is potential energy.
|
||
此时从位置a扔一个皮球,假设皮球只能落在位置a,b,c,而且皮球将等可能地落在这三个位置上。 | ||
|
||
系统处于a,b等位置的静态概率 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
这里是在定义静态概率吗?
由于 | ||
$$1-p(a\to a)=p(a \to b)+p(a \to c)$$ | ||
|
||
代入上式,得到 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Here it contains a typo in the main paper.
alpha = min(1,np.exp(U(x0) - U(xStar))) | ||
if alpha < rand(): | ||
xStar = x0 | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
最后的时候列一下参考文献~
@yufree 作者是不太能看到我上面的评论吗,还是有其他困难 |
我也奇怪,过会我发邮件问下。 |
@yufree 有后续吗 |
邮件当天就发了,但木有回音,我再问问。 |
@yufree 好像没有后续了 |
@statisticalcomputing 有时间修改一下文章吗?主站投稿应从作者最后一次回复后的三个月内处理完成,如果感觉时间不够,到10月6日后我们可以先暂时关闭,待有时间后可联系编辑部重新打开 PR 继续修改。请确认,谢谢! |
作者已联系编辑部,目前比较繁忙,但会有后续修改及投稿 @zhanruicai |
pull request, equation did not display,
original blog: https://statisticalcomputing.github.io/hmc1.html
非常感谢您的PR, 如果您是在为主站投稿, 请将PR的标题改为"投稿:标题+作者的形式",如:
"投稿: 数据通灵术 杜亚磊"
并保留下面的内容.
forum_id
加入文章投稿指南在这里,有任何问题,可以直接在PR留言,其他问题请联系: editor@cos.name。