<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[bofu_sun的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/bofu_sun</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; bofu_sun]]></copyright><item><title><![CDATA[分层强化学习 综述论文阅读 Hierarchical Reinforcement Learning: A Comprehensive Survey]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/133469012</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/133469012</guid><author>bofu_sun</author><pubDate>Sat, 07 Oct 2023 17:05:19 +0800</pubDate><description><![CDATA[分层强化学习可以通过将困难的长期决策任务分解为更简单的子任务，提升强化学习算法的性能。分层强化学习方法主要涉及：使用HRL学习分层策略、子任务发现、迁移学习和多智能体学习四个主要挑战。强化学习算法的一个痛点：如果任务的长度很长，状态空间和动作空间很大，由于难以探索，强化学习算法很难获得好的效果。分层强化学习的优点：分层强化学习将整个任务分成多个子任务。子任务本身更容易学习，所学习的子任务会导致更结构化的探索。分层强化学习效果好的原因：HRL在长期任务中效果优于标准RL，其优势在于改进了探索。]]></description><category></category></item><item><title><![CDATA[强化学习泛化性 论文阅读 Invariant Causal Imitation Learning for Generalizable Policies]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/128909829</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/128909829</guid><author>bofu_sun</author><pubDate>Tue, 07 Feb 2023 14:43:47 +0800</pubDate><description><![CDATA[论文阅读  Invariant Causal Imitation Learning for Generalizable Policies]]></description><category></category></item><item><title><![CDATA[强化学习泛化性 综述论文阅读 A SURVEY OF GENERALISATION IN DEEP REINFORCEMENT LEARNING]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/128076452</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/128076452</guid><author>bofu_sun</author><pubDate>Tue, 29 Nov 2022 23:43:26 +0800</pubDate><description><![CDATA[论文阅读]]></description><category></category></item><item><title><![CDATA[git 上传代码]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/128058007</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/128058007</guid><author>bofu_sun</author><pubDate>Sun, 27 Nov 2022 14:48:04 +0800</pubDate><description><![CDATA[git 上传代码]]></description><category></category></item><item><title><![CDATA[论文阅读 Fast Reinforcement Learning Via Slow Reinforcement Learning]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/128020981</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/128020981</guid><author>bofu_sun</author><pubDate>Thu, 24 Nov 2022 16:50:23 +0800</pubDate><description><![CDATA[论文阅读 元强化学习 RL^2]]></description><category></category></item><item><title><![CDATA[元强化学习 论文理解 MAESN]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/127879443</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/127879443</guid><author>bofu_sun</author><pubDate>Wed, 16 Nov 2022 10:42:28 +0800</pubDate><description><![CDATA[MAESN论文阅读]]></description><category></category></item><item><title><![CDATA[论文阅读 MAML (Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks)]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/127844484</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/127844484</guid><author>bofu_sun</author><pubDate>Mon, 14 Nov 2022 23:03:12 +0800</pubDate><description><![CDATA[元学习论文MAML阅读]]></description><category></category></item><item><title><![CDATA[元学习基础理解]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/127841270</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/127841270</guid><author>bofu_sun</author><pubDate>Mon, 14 Nov 2022 10:14:42 +0800</pubDate><description><![CDATA[元学习基础理解]]></description><category></category></item><item><title><![CDATA[离线强化学习论文学习 Critic Regularized Regression]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/126712305</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/126712305</guid><author>bofu_sun</author><pubDate>Wed, 07 Sep 2022 16:21:27 +0800</pubDate><description><![CDATA[CRR (critic regularized regression) 离线强化学习算法论文阅读总结]]></description><category></category></item><item><title><![CDATA[服务器传输大文件方法]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/125249502</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/125249502</guid><author>bofu_sun</author><pubDate>Sun, 12 Jun 2022 20:35:08 +0800</pubDate><description><![CDATA[从服务器下载大文件方法]]></description><category></category></item><item><title><![CDATA[2021-09-02 在idea中配置maven时，导入本地包]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/120065180</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/120065180</guid><author>bofu_sun</author><pubDate>Thu, 02 Sep 2021 16:57:31 +0800</pubDate><description><![CDATA[在idea中配置maven时报错：
Could not find artifact com.xxxx:bikini:pom:1.0.0 in tencent_spark_repo (https://mirrors.tencent.com/repository/maven/tencent_spark_release)

原因是网上确实缺失这个依赖，因此需要从网上找到jar包，下载到本地后手动导入。
在cmd中运行
mvn install:install-file -DgroupId=com.xxxx -Dar]]></description><category></category></item><item><title><![CDATA[论文阅读 Solving Mixed Integer Programs Using Neural Networks]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/119852854</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/119852854</guid><author>bofu_sun</author><pubDate>Mon, 23 Aug 2021 10:19:14 +0800</pubDate><description><![CDATA[论文：用神经网络求解混合整数规划
1.摘要
（1）文章目的：
混合整数规划（MIP）求解器一般使用一系列启发式算法来求解。而机器学习算法能		够通过利用数据实例之间的共享结构，构造更好的启发式算法。
（2）文章工作：
将学习算法应用于 Diving 和 Branching 两个子任务，即 Neural Diving 和 Neural Branching。其中 Neural Diving 即使用神经网络为整型变量生成部分的解，其余的变量使用Scip求解器求解（个人理解是为了防止决策变量过多的问题）。其中 Ne]]></description><category></category></item><item><title><![CDATA[多gpu训练]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/119804268</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/119804268</guid><author>bofu_sun</author><pubDate>Thu, 19 Aug 2021 16:05:35 +0800</pubDate><description><![CDATA[最近在使用多gpu进行训练，加入一下几点即可

加入使用的gpu

os.environ['CUDA_VISIBLE_DEVICES'] = '0,1,2,3,4,5,6,7'
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


将模型加载到设备上

self.actor_net = nn.DataParallel(Mlp1_Actor_model(input_dim).to(device=device))


将数]]></description><category></category></item><item><title><![CDATA[output = input.matmul(weight.t())      RuntimeError: expected scalar type Long but found Float 错误解决]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/117068289</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/117068289</guid><author>bofu_sun</author><pubDate>Thu, 20 May 2021 11:19:38 +0800</pubDate><description><![CDATA[在使用pytorch的nn.Linear时出现错误
 RuntimeError: expected scalar type Long but found Float

这里报错的原因是我的输入是Longtensor，因此这里将输入改为Floattensor数据类型就可以了。

]]></description><category></category></item><item><title><![CDATA[pytorch常用工具小结]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/116743859</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/116743859</guid><author>bofu_sun</author><pubDate>Tue, 18 May 2021 19:40:20 +0800</pubDate><description><![CDATA[getitem 方法
在类中有__getitem__ 方法，代表当实例对象通过[] 运算符取值时，会调用它的方法__getitem__。
torch.utils.data.DataLoader
参数：
(1) dataset: 使用的数据集。
(2) batch_size：每个batch的数量。
(3) shuffle: 是否打乱顺序。
(4) sampler: 采样方法，和shuffle互斥。
(5) batch_sampler：批量采样，和batch_size、shuffle互斥。
(7) num_w.]]></description><category></category></item><item><title><![CDATA[python 频率分布直方图]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/109107671</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/109107671</guid><author>bofu_sun</author><pubDate>Thu, 15 Oct 2020 23:36:35 +0800</pubDate><description><![CDATA[在处理特征的过程中，往往在分析对比数据时需要分析数据的分布情况，这时在python中常用到matplotlib中的pyplot模块中的hist函数。
但在使用hist时存在着如果设置normed，改变bins数量会导致概率变化的问题，因此在绘制分布直方图时需要将bins数量设置为真实的数量。
2.hist参数解析
matplotlib.pyplot.hist(
  x, bins=10, range=None, normed=False,
  weights=None, cumulative=Fals...]]></description><category></category></item><item><title><![CDATA[相关系数]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/109104936</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/109104936</guid><author>bofu_sun</author><pubDate>Thu, 15 Oct 2020 22:12:51 +0800</pubDate><description><![CDATA[在处理数据关系，建立模型时，我们往往会先对于可能影响目标的特征进行筛选，因为并不是所有的特征都对于我们的目标存在很大的影响。
1.对于相关系数有如下的理解：
如果有两个变量：X、Y，最终计算出的相关系数的含义可以有如下理解：
(1)、当相关系数为0时，X和Y两变量无关系。
(2)、当X的值增大（减小），Y值增大（减小），两个变量为正相关，相关系数在0.00与1.00之间。
(3)、当X的值增大（减小），Y值减小（增大），两个变量为负相关，相关系数在-1.00与0.00之间。
相关系数的绝对值越大，相关性越强]]></description><category></category></item><item><title><![CDATA[docker已被占用解决方法]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/108229656</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/108229656</guid><author>bofu_sun</author><pubDate>Tue, 25 Aug 2020 21:47:55 +0800</pubDate><description><![CDATA[docker已被占用解决方法
(my_python3) sun@sun-Inspiron-3543:~/project/my_realize/huawei/starter_kit$ docker ps -a
CONTAINER ID        IMAGE               COMMAND             CREATED             STATUS              PORTS               NAMES
0c9f7985619a        smarts]]></description><category></category></item><item><title><![CDATA[torchvision安装填坑]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/106968946</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/106968946</guid><author>bofu_sun</author><pubDate>Fri, 26 Jun 2020 13:54:44 +0800</pubDate><description><![CDATA[这两天安装torchvision出现了很多坑，这里记录一下。
开始我的torch版本是1.3.1，正常pip安装torchvision
pip install torchvision

安装成功，但是导入的时候出现了问题。
&gt;&gt;&gt; import torchvision
Traceback (most recent call last):
  File "&lt;stdin&gt;", line 1, in &lt;module&gt;
  File "/home/sun/my_python]]></description><category></category></item><item><title><![CDATA[maddpg Gumbel-Softmax Trick]]></title><link>https://blog.youkuaiyun.com/bofu_sun/article/details/106697798</link><guid>https://blog.youkuaiyun.com/bofu_sun/article/details/106697798</guid><author>bofu_sun</author><pubDate>Thu, 11 Jun 2020 21:50:56 +0800</pubDate><description><![CDATA[https://www.cnblogs.com/initial-h/p/9468974.html

]]></description><category></category></item></channel></rss>