<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[DS_agent的博客]]></title><description><![CDATA[]]></description><link>https://blog.csdn.net/DS_agent</link><language>zh-cn</language><generator>https://blog.csdn.net/</generator><copyright><![CDATA[Copyright &copy; DS_agent]]></copyright><item><title><![CDATA[基于PyTorch的联邦学习开源框架FedLab：加速FL算法验证]]></title><link>https://blog.csdn.net/DS_agent/article/details/119792845</link><guid>https://blog.csdn.net/DS_agent/article/details/119792845</guid><author>DS_agent</author><pubDate>Thu, 19 Aug 2021 02:25:09 +0800</pubDate><description><![CDATA[FL框架调研
近两年来联邦学习federated learning（FL）非常火热，也有很多开源的联邦学习框架。

微众银行的FATE

FATE文档
覆盖横向、纵向、迁移联邦学习；实现了样本安全匹配，样本切分、特征处理和筛选、LR/XGB/DNN等常用算法，模型评估与评分卡，模型预测（serving），联邦推荐等；文档丰富
工业级框架，比较繁重，上手比较困难，不易于实验室环境部署


OpenMind的PySyft: 由openmind大力推广，PySyft 在主要深度学习框架（如 PyTorch 和 .]]></description><category></category></item><item><title><![CDATA[Ubuntu/Linux下自动获取最佳GPU编号的脚本]]></title><link>https://blog.csdn.net/DS_agent/article/details/119616245</link><guid>https://blog.csdn.net/DS_agent/article/details/119616245</guid><author>DS_agent</author><pubDate>Wed, 11 Aug 2021 21:08:20 +0800</pubDate><description><![CDATA[PyTorch中如果直接用方法tensor.cuda()或tensor.device("cuda")将变量放到gpu上时默认将数据放到cuda:0上，而没有办法选择拥有最大空余存储的GPU上。因此写了一个脚本用来在Linux系统下自动获取拥有最大空余存储的GPU的对应ID（假设4张卡，默认ID为0到3）。
系统要求：Linux系统，且安装了nvidia-smi
首先给出实验设备的系统和环境：
# 查看操作系统
$ head -n 1 /etc/issue
Ubuntu 18.04.5 LTS \n \l

]]></description><category></category></item><item><title><![CDATA[Pytorch学习笔记-Autograd自动求导与梯度操作]]></title><link>https://blog.csdn.net/DS_agent/article/details/117881073</link><guid>https://blog.csdn.net/DS_agent/article/details/117881073</guid><author>DS_agent</author><pubDate>Sun, 13 Jun 2021 16:52:00 +0800</pubDate><description><![CDATA[tensor.grad和tensor.grad.data区别是什么？
手动进行梯度更新时，如果直接用tensor.grad进行运算则会被加入计算图，导致内存占用不停加大以及累计梯度出错，用tensor.grad.data则不会。或者可以在with torch.no_grad():下直接用tensor.grad进行，这样就不会被加入计算图或者梯度累加

__注意：__不管哪种方法，每次更新完都记得对梯度清空！！！



Pytorch教程中有一段关于自定义autograd函数的代码
class Legen..]]></description><category></category></item><item><title><![CDATA[Pytorch学习笔记-Basic Data Type基本数据类型]]></title><link>https://blog.csdn.net/DS_agent/article/details/117855178</link><guid>https://blog.csdn.net/DS_agent/article/details/117855178</guid><author>DS_agent</author><pubDate>Sat, 12 Jun 2021 20:27:49 +0800</pubDate><description><![CDATA[少用tensor.numpy()来查看tensor的np.ndarray格式数据，共享内存，修改torch.tensor对应的np.ndarray也会被改动


少用torch.from_numpy()来生成torch.tensor，共享内存，修改原始np.ndarray，其生成的torch.tensor也会被修改


如果是希望防止跟踪历史（以及使用内存），可以将代码块放在 with torch.no_grad(): 内，这个做法在使用一个模型进行评估的时候非常有用，因为模型会包含一些带有 requi..]]></description><category></category></item><item><title><![CDATA[LASSO近端梯度下降法Proximal Gradient Descent公式推导及代码]]></title><link>https://blog.csdn.net/DS_agent/article/details/106340777</link><guid>https://blog.csdn.net/DS_agent/article/details/106340777</guid><author>DS_agent</author><pubDate>Mon, 25 May 2020 20:05:00 +0800</pubDate><description><![CDATA[文章目录LASSO by Proximal Gradient DescentProximal Gradient Descent FrameworkProximal Gradient Descent DetailsSimplified Code
LASSO by Proximal Gradient Descent
Proximal Gradient Descent Framework

randomly set β(0)\beta^{(0)}β(0) for iteration 0
For kkkth ite]]></description><category></category></item><item><title><![CDATA[LASSO坐标下降法Coordinate Descent Method公式推导及代码]]></title><link>https://blog.csdn.net/DS_agent/article/details/106340309</link><guid>https://blog.csdn.net/DS_agent/article/details/106340309</guid><author>DS_agent</author><pubDate>Mon, 25 May 2020 19:43:34 +0800</pubDate><description><![CDATA[文章目录LASSO by Coordinate Descent MethodCoordinate Descent Method FrameworkCoordinate Descent Method Update DetailsSimplified CodeMy LASSO Path
LASSO by Coordinate Descent Method
Prepare:
from itertools import cycle
import numpy as np
import matplotlib.pyplo]]></description><category></category></item><item><title><![CDATA[组合优化中的全排列生成方法]]></title><link>https://blog.csdn.net/DS_agent/article/details/62056849</link><guid>https://blog.csdn.net/DS_agent/article/details/62056849</guid><author>DS_agent</author><pubDate>Tue, 14 Mar 2017 21:17:02 +0800</pubDate><description><![CDATA[组合优化中的全排列生成之前有同学遇到组合优化（如0-1优化）问题，想采用穷举法，那么首先就要穷举产生所有的组合。以0-1优化为例，假设当前有3个item，每个item有“选”或“不选”两种状态。]]></description><category></category></item><item><title><![CDATA[Tensor Decompositions and Applications第一、二章翻译]]></title><link>https://blog.csdn.net/DS_agent/article/details/53154280</link><guid>https://blog.csdn.net/DS_agent/article/details/53154280</guid><author>DS_agent</author><pubDate>Mon, 14 Nov 2016 02:04:06 +0800</pubDate><description><![CDATA[《张量分解与应用》第一二章翻译：张量是一种多维数组。一个一阶张量就是一个向量，一个二阶张量是一个矩阵，具有三阶或更多阶数的张量称为高阶数组。张量的阶（order）即张量的维数（dimension），也称为way或mode。  
    向量（一阶张量）用粗体小写字母表示，如. 
    矩阵（二阶张量）用粗体大写字母表示，如.]]></description><category></category></item><item><title><![CDATA[Matrix and Tensor Decomposition in Recommender Systems 阅读笔记（翻译）]]></title><link>https://blog.csdn.net/DS_agent/article/details/52802450</link><guid>https://blog.csdn.net/DS_agent/article/details/52802450</guid><author>DS_agent</author><pubDate>Thu, 13 Oct 2016 00:29:37 +0800</pubDate><description><![CDATA[Matrix and Tensor Decomposition in RecommenderSystems
阅读笔记（翻译）

2.MATRIX DECOMPOSITION
    矩阵分解（matrix factorization）是将一个矩阵分解为多个矩阵相乘结果的过程，它对发掘参与实体（participating entities, 如用户和商品）的数据中的潜在关系有着重要意义。在简]]></description><category></category></item><item><title><![CDATA[UVa1339 Ancient Cipher]]></title><link>https://blog.csdn.net/DS_agent/article/details/52745679</link><guid>https://blog.csdn.net/DS_agent/article/details/52745679</guid><author>DS_agent</author><pubDate>Thu, 06 Oct 2016 21:30:23 +0800</pubDate><description><![CDATA[原题：
Ancient Roman empire had a strong government system with various departments, including a secretservice department. Important documents were sent between provinces and the capital in
 encryptedf]]></description><category></category></item><item><title><![CDATA[UVa227 Puzzle]]></title><link>https://blog.csdn.net/DS_agent/article/details/52744701</link><guid>https://blog.csdn.net/DS_agent/article/details/52744701</guid><author>DS_agent</author><pubDate>Thu, 06 Oct 2016 17:41:53 +0800</pubDate><description><![CDATA[这题的坑在于输出格式。。。
原题：
A children’s puzzle that was popular 30 years ago consisted of a 5×5 frame which contained 24 smallsquares of equal size. A unique letter of the alphabet was printed on each small ]]></description><category></category></item><item><title><![CDATA[UVa455 Periodic Strings]]></title><link>https://blog.csdn.net/DS_agent/article/details/52740654</link><guid>https://blog.csdn.net/DS_agent/article/details/52740654</guid><author>DS_agent</author><pubDate>Wed, 05 Oct 2016 21:20:03 +0800</pubDate><description><![CDATA[原题：
A character string is said to have period k if it can be formed by concatenating one or more repetitionsof another string of length
k. For example, the string ”abcabcabcabc” has period 3, since ]]></description><category></category></item><item><title><![CDATA[UVa1225 Digit Counting]]></title><link>https://blog.csdn.net/DS_agent/article/details/52740291</link><guid>https://blog.csdn.net/DS_agent/article/details/52740291</guid><author>DS_agent</author><pubDate>Wed, 05 Oct 2016 20:11:01 +0800</pubDate><description><![CDATA[原题：

Trung is bored with his mathematics homeworks. He takes a piece of chalk and starts writing a sequence of consecutive integers starting with 1 to N (1
 N  . After that, he counts the number of]]></description><category></category></item><item><title><![CDATA[UVa1584 Circular Sequence]]></title><link>https://blog.csdn.net/DS_agent/article/details/52739760</link><guid>https://blog.csdn.net/DS_agent/article/details/52739760</guid><author>DS_agent</author><pubDate>Wed, 05 Oct 2016 17:46:33 +0800</pubDate><description><![CDATA[Some DNA sequences exist in circular forms as in the following figure, which shows a circular sequence ``CGAGTCAGCT", that is, the last symbol ``T" in ``CGAGTCAGCT" is connected to the first symbol ``C". We always read a circular sequence in the clockwise ]]></description><category></category></item><item><title><![CDATA[UVa1585 Score]]></title><link>https://blog.csdn.net/DS_agent/article/details/52738850</link><guid>https://blog.csdn.net/DS_agent/article/details/52738850</guid><author>DS_agent</author><pubDate>Wed, 05 Oct 2016 15:12:19 +0800</pubDate><description><![CDATA[原题：

There is an objective test result such as ``OOXXOXXOOO". An `O' means a correct answer of a problem and an `X' means a wrong answer. The score of each problem
 of this test is calculated by it]]></description><category></category></item><item><title><![CDATA[USA1586 Molar mass]]></title><link>https://blog.csdn.net/DS_agent/article/details/52738824</link><guid>https://blog.csdn.net/DS_agent/article/details/52738824</guid><author>DS_agent</author><pubDate>Wed, 05 Oct 2016 15:06:32 +0800</pubDate><description><![CDATA[原题：




代码：
#include 
#include 
#include 
#define maxn 80
char chem[maxn];
int main() {
    using namespace std;
    int T, n;
    int i;
    double mass, per;
    char c;
    if (scanf("%d",&T)]]></description><category></category></item><item><title><![CDATA[UVa401 Palindromes]]></title><link>https://blog.csdn.net/DS_agent/article/details/52730826</link><guid>https://blog.csdn.net/DS_agent/article/details/52730826</guid><author>DS_agent</author><pubDate>Tue, 04 Oct 2016 02:08:47 +0800</pubDate><description><![CDATA[原题：输入一个字符串，判断它是否为回文串以及镜像串。输入字符串保证不含数字0。所谓回文串，即反转后和原串相同，如abba和madam。所谓镜像串, 是指左右镜像后原串相同,如 2S 和 3AIAE. 注意,并不是所有的字符在镜像之后都会得到一个合法字符。每个字符的镜像如下图所示：





Character
Reverse
Character
Reverse
Characte]]></description><category></category></item><item><title><![CDATA[《機器學習實戰》學習筆記]]></title><link>https://blog.csdn.net/DS_agent/article/details/51702133</link><guid>https://blog.csdn.net/DS_agent/article/details/51702133</guid><author>DS_agent</author><pubDate>Sat, 18 Jun 2016 00:09:15 +0800</pubDate><description><![CDATA[對機器學習感興趣的大二黨，暑假過後就要升大三了，學姊和導師推薦過《機器學習實戰》。
原來已經接觸Matlab快大半年了（被學姊吐槽CS），而《機器學習實戰》裡面用的是python，所以同時自學python，不是很熟悉。
在這裡主要是想寫一些總結，和自己用Matlab實現的代碼]]></description><category></category></item><item><title><![CDATA[pagerank算法的Matlab實現]]></title><link>https://blog.csdn.net/DS_agent/article/details/50539635</link><guid>https://blog.csdn.net/DS_agent/article/details/50539635</guid><author>DS_agent</author><pubDate>Mon, 18 Jan 2016 22:43:39 +0800</pubDate><description><![CDATA[网页搜索中的网页排名PageRank( L, sigma )
 基本转移矩阵M满足: M= L' * D^(-1)     上标一撇'表示转秩
 L为网络图中的邻接矩阵,D为对角线上元素D(k,k)表示第k个点出度的对角矩阵
 处理悬挂网页的随机性修正: S= M + e*a'/N
 e为所有分量均为1的列向量, N为网页总数
 a为用于描述“悬挂网页”的行向量, 其第i个分量的取值由第]]></description><category></category></item><item><title><![CDATA[《數據結構與算法分析：C語言描述》習題2.11代碼及自己的想法]]></title><link>https://blog.csdn.net/DS_agent/article/details/47685727</link><guid>https://blog.csdn.net/DS_agent/article/details/47685727</guid><author>DS_agent</author><pubDate>Sat, 15 Aug 2015 21:04:05 +0800</pubDate><description><![CDATA[《數據結構與算法分析：C語言描述》習題2.11
大一狗放假在家自學點算法，編程渣，自己練習一下
————————————————————————————————（我是劃分線）
題目：给出一个有效算法来确定整数A[1]的数组中是否存在整数i使得A[i]=i.
分析：用二分查找法的變體即可
上代碼：
#include 
int Search(int A[], int Left, int R]]></description><category></category></item></channel></rss>