<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[PuJiang-的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/jump882</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; jump882]]></copyright><item><title><![CDATA[sigmoid、softmax函数/交叉熵损失函数/二分类、多分类详细推导过程]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/126492348</link><guid>https://blog.youkuaiyun.com/jump882/article/details/126492348</guid><author>jump882</author><pubDate>Tue, 23 Aug 2022 22:05:54 +0800</pubDate><description><![CDATA[交叉熵损失函数、MSE、二分类、多分类]]></description><category></category></item><item><title><![CDATA[线性回归(Linear Regression)、逻辑回归(Logistic Regression)的原理、联系与区别]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/126490003</link><guid>https://blog.youkuaiyun.com/jump882/article/details/126490003</guid><author>jump882</author><pubDate>Tue, 23 Aug 2022 18:00:31 +0800</pubDate><description><![CDATA[线性回归、逻辑回归、原理]]></description><category></category></item><item><title><![CDATA[逻辑回归(Logistic Regression)原理及损失函数、极大似然估计]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/126485676</link><guid>https://blog.youkuaiyun.com/jump882/article/details/126485676</guid><author>jump882</author><pubDate>Tue, 23 Aug 2022 16:43:36 +0800</pubDate><description><![CDATA[逻辑回归原理及公式推导过程]]></description><category></category></item><item><title><![CDATA[模型评估指标_精准率、召回率、F1-score、ROC曲线、AUC曲线]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/124583868</link><guid>https://blog.youkuaiyun.com/jump882/article/details/124583868</guid><author>jump882</author><pubDate>Thu, 05 May 2022 12:51:13 +0800</pubDate><description><![CDATA[一、分类结果混淆矩阵

1、T、F、P、N辨析

P(Positive)、N(Negative)针对的是预测值的结果。
P表示对样本的预测结果为正例，
N表示对样本的预测结果为反例。
T(True)、F(False)针对的是预测值的结果是否预测对了。
TP表示对该样本预测结果为正例，同时预测对了，真实标签和预测标签都为正例，
FP表示对该样本预测结果为正例，但是预测错了，真实标签为反例，预测标签为正例，
TN表示对该样本预测结果为反例，同时预测对了，真是标签和预测标签都为反例，
FN表示对该样本预测结果为反]]></description><category></category></item><item><title><![CDATA[Python英文分词_词频统计_按词频排序]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/124326361</link><guid>https://blog.youkuaiyun.com/jump882/article/details/124326361</guid><author>jump882</author><pubDate>Thu, 21 Apr 2022 17:10:42 +0800</pubDate><description><![CDATA[文本cet6read.txt

QuestionS 61 to 65 are based on the following passage.
Many bankers may be worried about whether some fancy product dreamed up might yetlead to a visit from the police. Daniel Dantas, a financier, who has profited by operating at themiddle ]]></description><category></category></item><item><title><![CDATA[Python数据处理_pandas读取csv文件_numpy进行归一化处理]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/124322499</link><guid>https://blog.youkuaiyun.com/jump882/article/details/124322499</guid><author>jump882</author><pubDate>Thu, 21 Apr 2022 15:32:38 +0800</pubDate><description><![CDATA[请执行以下命令下载样本数据
数据为CSV表格有8列，其中abstime列为时间列，其余列为特征列
!wget http://82.156.209.173/s/dHCnaH32rtZdxcs/download
!mv download sample_data_A100159.csv

1. 对给出的样本文件中的所有特征列做normalization，映射至（0-255）的区间，从abstime的第一行开始每隔200条取一次整行数据，并将最终取出的数据并以PNG格式的图片输出。

提示：输出图片用PIL库
可参]]></description><category></category></item><item><title><![CDATA[Python内置数据结构_元组、列表、字典_基本操作]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/124320164</link><guid>https://blog.youkuaiyun.com/jump882/article/details/124320164</guid><author>jump882</author><pubDate>Thu, 21 Apr 2022 14:48:56 +0800</pubDate><description><![CDATA[一、Python元组
Python 的元组与列表类似，不同之处在于元组的元素不能修改。

元组使用小括号，列表使用中括号
元组不允许修改元素值，只能进行多个元组的连接组合，这样生成的是一个全新的元组
元组不允许删除元素值，只能删除整个元组

元组进行的都是整体级别的操作，内部细粒度的元素不能更改，首次创建好之后进行的都是元组整体的操作了。
1.1元组创建
tup1 = ('physics', 'chemistry', 1997, 2000)
tup2 = (1, 2, 3, 4, 5 )
tup3 = "a]]></description><category></category></item><item><title><![CDATA[Attention的汇总与辨析_Additive、Multiplication、Scaled dot-product、Self Attention、Multi-head Self-Attention]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/123847336</link><guid>https://blog.youkuaiyun.com/jump882/article/details/123847336</guid><author>jump882</author><pubDate>Thu, 07 Apr 2022 17:08:52 +0800</pubDate><description><![CDATA[一、Seq2Seq
1.1Seq2Seq(Encoder-Decoder)是什么

简介：使用Encoder将input编码为一个固定长度的context向量，使用Decoder将context解码为output。input、output长度不一定相同。
奠基论文：Sequence to Sequence Learning with Neural Networks



...]]></description><category></category></item><item><title><![CDATA[RNN改进_GRU_LSTM基本原理及辨析、GRU激活函数为什么使用Sigmoid而不是ReLU]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/123741482</link><guid>https://blog.youkuaiyun.com/jump882/article/details/123741482</guid><author>jump882</author><pubDate>Mon, 28 Mar 2022 16:09:43 +0800</pubDate><description><![CDATA[一、RNN存在的问题
对RNN来讲，没有特别关注的部分，都是一个序列进去。而且RNN的梯度消失指的是每一项一项进行相加，可以发现距离拉的越长，连乘的项就越多，远距离的梯度会趋于0的，近距离的梯度不会消失。RNN梯度消失的真正含义是总的梯度受近距离梯度的主导，远距离的梯度消失。
二、GRU门控单元
引入重置门和更新门：
Rt=σ(XtWwr+Ht−1Whr+br)Zt=σ(XtWwz+Ht−1Whz+bz)R_t=\sigma(X_tW_{wr}+H_{t-1}W_{hr}+b_r) \\ Z_t=\sigm]]></description><category></category></item><item><title><![CDATA[深度神经网络中的激活函数_导数推理过程_python绘制图像示意_特点及使用场景]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/123750100</link><guid>https://blog.youkuaiyun.com/jump882/article/details/123750100</guid><author>jump882</author><pubDate>Sat, 26 Mar 2022 13:37:03 +0800</pubDate><description><![CDATA[1、Sigmoid激活函数
f(z)=11+e−zf(z)=\frac{1}{1+e^{-z}}f(z)=1+e−z1​f′(z)=0−1∗(1+e−z)′(1+e−z)2=−e−z∗−1(1+e−z)2=e−z(1+e−z)2=1+e−z−1(1+e−z)2=1(1+e−z)−1(1+e−z)2=f(z)(1−f(z)) f^{'}(z)=\frac{0-1*(1+e^{-z})^{'}}{(1+e^{-z})^2}\\ =\frac{-e^{-z}*-1}{(1+e^{-z})^2}\\=\frac{e]]></description><category></category></item><item><title><![CDATA[RNN基本原理及梯度消失、梯度爆炸的问题原因及解决方法]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/123717798</link><guid>https://blog.youkuaiyun.com/jump882/article/details/123717798</guid><author>jump882</author><pubDate>Fri, 25 Mar 2022 11:09:24 +0800</pubDate><description><![CDATA[一、RNN基本结构

1、隐层状态sts_tst​
st=σ(Uxt+Wst−1+b)s_t=\sigma(Ux_t+Ws_{t-1}+b)st​=σ(Uxt​+Wst−1​+b)
σ\sigmaσ()是激活函数，通常选用Tanh、ReLU。
2、输出状态oto_tot​
ot=g(Vst+c)o_t=g(Vs_t+c)ot​=g(Vst​+c)
ggg()是激活函数，对于分类任务通常选用sigmoidsigmoidsigmoid()。
3、Loss计算
输出状态oto_tot​与目标输出yty_tyt​计]]></description><category></category></item><item><title><![CDATA[Pytorch计算Loss值为Nan的一种情况【exp计算溢出，利用softmax计算的冗余性解决】]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/121371018</link><guid>https://blog.youkuaiyun.com/jump882/article/details/121371018</guid><author>jump882</author><pubDate>Wed, 17 Nov 2021 10:49:20 +0800</pubDate><description><![CDATA[一、报错提示
FutureWarning: Non-finite norm encountered in torch.nn.utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_n]]></description><category></category></item><item><title><![CDATA[beam search]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/120001209</link><guid>https://blog.youkuaiyun.com/jump882/article/details/120001209</guid><author>jump882</author><pubDate>Mon, 30 Aug 2021 17:12:02 +0800</pubDate><description><![CDATA[]]></description><category></category></item><item><title><![CDATA[Linux操作系统安全登录实现_usb认证安全加固]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/119881732</link><guid>https://blog.youkuaiyun.com/jump882/article/details/119881732</guid><author>jump882</author><pubDate>Tue, 24 Aug 2021 11:21:19 +0800</pubDate><description><![CDATA[一、实验目的
本实验在linux系统中使用usb进行身份验证。采用双重登陆方法即输入正确的登陆密码同时还需要插入正确的usb设备才能通过身份认证，以此来实现linux系统登录过程中的安全加固。
二、实验环境
1、系统环境
虚拟机版本：VMware Workstation 16 Pro 16.0.0
Linux版本：2.6.32-131.0.15.el6.x86_64
gcc版本：4.4.4 20100726 (Red Hat 4.4.4-13) (GCC)
发行版本：CentOS 6.1
2、插件及依赖包
]]></description><category></category></item><item><title><![CDATA[tensorboard_SummaryWriter_event定义及使用示例]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/119862224</link><guid>https://blog.youkuaiyun.com/jump882/article/details/119862224</guid><author>jump882</author><pubDate>Mon, 23 Aug 2021 10:20:22 +0800</pubDate><description><![CDATA[一、导包并创建SummaryWriter
使用SummaryWriter创建events：
from torch.utils.tensorboard import SummaryWriter

writer = SummaryWriter()
writer = SummaryWriter("test0")
writer = SummaryWriter(comment="user")

SummaryWriter()：自动生成文件夹名称"runs"，以及随机的events名称
SummaryWriter(“t]]></description><category></category></item><item><title><![CDATA[Scaled dot-product Attention、Self-Attention辨析]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/119852331</link><guid>https://blog.youkuaiyun.com/jump882/article/details/119852331</guid><author>jump882</author><pubDate>Sun, 22 Aug 2021 15:50:50 +0800</pubDate><description><![CDATA[一、Scaled dot-product Attention
有两个序列X、YX、YX、Y：序列XXX提供查询信息QQQ，序列YYY提供键、值信息K、VK、VK、V。Q∈Rx_len×in_dimQ\in R^{{x\_len}\times {in\_dim}}{}Q∈Rx_len×in_dimK∈Ry_len×in_dimK\in R^{{y\_len}\times {in\_dim}}{}K∈Ry_len×in_dimV∈Ry_len×out_dimV\in R^{{y\_len}\times {out]]></description><category></category></item><item><title><![CDATA[Self-Attention原理、Multi-head Self-Attention原理及Pytorch实现]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/119799992</link><guid>https://blog.youkuaiyun.com/jump882/article/details/119799992</guid><author>jump882</author><pubDate>Thu, 19 Aug 2021 17:43:12 +0800</pubDate><description><![CDATA[一、Self-Attention原理
下图虚线框内为Self-Attention模块基本功能，输入a1,a2,a3,a4a_1,a_2,a_3,a_4a1​,a2​,a3​,a4​，输出b1,b2,b3,b4b_1,b_2,b_3,b_4b1​,b2​,b3​,b4​。输入序列与输出序列长度相同，内部实际上做的是加权求和的运算。

1、计算a1a_1a1​与所有向量(包括自己)的attention-score
使用q1q_1q1​与k1,k2,k3,k4k_1,k_2,k_3,k_4k1​,k2​,k3​,]]></description><category></category></item><item><title><![CDATA[BatchNorm、LayerNorm详细过程及示例_Pytorch]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/119795466</link><guid>https://blog.youkuaiyun.com/jump882/article/details/119795466</guid><author>jump882</author><pubDate>Thu, 19 Aug 2021 12:31:25 +0800</pubDate><description><![CDATA[一、BatchNorm
1、对batch中的每个维度进行归一化
y=x−E[x]Var[x]y=\frac{x-E[x]}{\sqrt{Var[x]}}y=Var[x]​x−E[x]​其中：E[x]=1n∑i=1nxi,Var[x]=1n∑i=1n(xi−E[x])2E[x]=\frac{1}{n}\sum_{i=1}^nx_i,Var[x]=\frac{1}{n}\sum_{i=1}^n(x_i-E[x])^2E[x]=n1​∑i=1n​xi​,Var[x]=n1​∑i=1n​(xi​−E[x])2

2]]></description><category></category></item><item><title><![CDATA[dropout手工实现_FashionMNIST数据集]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/119731645</link><guid>https://blog.youkuaiyun.com/jump882/article/details/119731645</guid><author>jump882</author><pubDate>Mon, 16 Aug 2021 15:49:06 +0800</pubDate><description><![CDATA[一、dropout原理
神经网络中dropout以p的概率丢弃该层的节点。p=0时，没有节点被丢弃。p=1时，将丢弃所有节点。被丢弃的节点hi=0h_i=0hi​=0,保留下来的节点hi=hi′h_i=h_i'hi​=hi′​。其中要满足E[hi′]=hiE[h_i']=h_iE[hi′​]=hi​。h′={0丢弃概率为ph1−p其他情况h'=
\begin{cases}
0&amp; \text{丢弃概率为p}\\
\frac{h}{1-p}&amp; \text{其他情况}
\end{cases}h′=]]></description><category></category></item><item><title><![CDATA[线性回归模型——范数、L1，L2正则化及其手工实现]]></title><link>https://blog.youkuaiyun.com/jump882/article/details/119728048</link><guid>https://blog.youkuaiyun.com/jump882/article/details/119728048</guid><author>jump882</author><pubDate>Mon, 16 Aug 2021 12:11:22 +0800</pubDate><description><![CDATA[一、p范数及实现
对于线性模型Y=XW+bY=XW+bY=XW+b，其中X∈Rn×dX\in R^{n \times d}X∈Rn×d，nnn为样本数，ddd为每个样本的特征维度，W∈Rd×1W \in R^{d \times 1}W∈Rd×1，Y∈Rn×1Y \in R^{n \times 1}Y∈Rn×1。可以使用权重向量WWW的某个范数来衡量该模型的复杂度。W=(w1,w2,...,wd)W =(w_1,w_2,...,w_d)W=(w1​,w2​,...,wd​)
1-范数：∣∣W∣∣1\vert\]]></description><category></category></item></channel></rss>