<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[dihe3772的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/dihe3772</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; dihe3772]]></copyright><item><title><![CDATA[GAN]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200553</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200553</guid><author>dihe3772</author><pubDate>Thu, 15 Aug 2019 13:36:00 +0800</pubDate><description><![CDATA[GAN，MNIST生成数字

import os
import matplotlib.pyplot as plt
import itertools
import pickle
import imageio
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.op...]]></description><category></category></item><item><title><![CDATA[ResNet-MNIST]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200543</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200543</guid><author>dihe3772</author><pubDate>Thu, 01 Aug 2019 13:17:00 +0800</pubDate><description><![CDATA[import torchvision
import torch
import torch.utils.data.dataloader as Data
from torch.autograd import Variable
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
fr...]]></description><category></category></item><item><title><![CDATA[Python显示Ground True]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200541</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200541</guid><author>dihe3772</author><pubDate>Tue, 30 Jul 2019 15:49:00 +0800</pubDate><description><![CDATA[数据使用Pascal voc 2007 里边的000012图像，将这张图片和对应的xml文件复制到代码同级目录下

import xml.etree.ElementTree as ET
import matplotlib.pyplot as plt
import matplotlib.image as mpimg # mpimg 用于读取图片
xml_path='000012....]]></description><category></category></item><item><title><![CDATA[PyTorch]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200540</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200540</guid><author>dihe3772</author><pubDate>Tue, 16 Jul 2019 18:41:00 +0800</pubDate><description><![CDATA[1.Torch.nn
class torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True)
卷积

//1是输入图像的channel
//6是输出图像的channel
//5是卷积核大小
nn.Conv2d(1, ...]]></description><category></category></item><item><title><![CDATA[EM算法]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200539</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200539</guid><author>dihe3772</author><pubDate>Wed, 22 May 2019 16:01:00 +0800</pubDate><description><![CDATA[转载自:https://www.cnblogs.com/Gabby/p/5344658.html
一、EM算法的预备知识
1、极大似然估计
（1）举例说明：经典问题——学生身高问题
　　我们需要调查我们学校的男生和女生的身高分布。 假设你在校园里随便找了100个男生和100个女生。他们共200个人。将他们按照性别划分为两组，然后先统计抽样得到的100个男生的身高。假设他们的身高是...]]></description><category></category></item><item><title><![CDATA[python基础]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200536</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200536</guid><author>dihe3772</author><pubDate>Wed, 22 May 2019 14:48:00 +0800</pubDate><description><![CDATA[1.数据类型
1.1数字类型
数字类型分为整数、浮点数、复数、布尔
整数默认十进制表示，二进制0B开头，八进制0O开头，十六进制0X开头
浮点数默认是float类型，可以使用科学技术法，1.5e-5表示$1.5 * 10^{-5}$
复数，1+2j，1是实部，2是虚部
布尔，True/False 2个值，可使用bool()函数将任意数据类型转换为bool型，空对象、列表、元组...]]></description><category></category></item><item><title><![CDATA[优化算法]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200534</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200534</guid><author>dihe3772</author><pubDate>Tue, 21 May 2019 22:07:00 +0800</pubDate><description><![CDATA[1.梯度下降法
1.1梯度
在微积分里面，对多元函数的参数求∂偏导数，把求得的各个参数的偏导数以向量的形式写出来，就是梯度。比如函数f(x,y), 分别对x,y求偏导数，求得的梯度向量就是(∂f/∂x,∂f/∂y)T,简称grad f(x,y)或者▽f(x,y)。
微积分中使用梯度表示函数增长最快的方向；因此，神经网络中使用负梯度来指示目标函数下降最快的方向，这样可以求得目标函数...]]></description><category></category></item><item><title><![CDATA[Faster R-CNN]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200532</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200532</guid><author>dihe3772</author><pubDate>Mon, 25 Mar 2019 10:11:00 +0800</pubDate><description><![CDATA[1.R-CNN
R-CNN网络架构图

R-CNN网络框架流程
1）原图像经过selective search算法提取约2000个候选框
2）候选框缩放到同一大小，原因是上图的ConvNet需要输入图片大小一致
3）通过ConvNet提取特征，原文ConvNet使用的是Alexnet，Alexnet需求的图片大小为（227*227），最后获得4096维特征向量
4）使用S...]]></description><category></category></item><item><title><![CDATA[SSD（single shot multibox detector）]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200526</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200526</guid><author>dihe3772</author><pubDate>Mon, 18 Mar 2019 15:53:00 +0800</pubDate><description><![CDATA[SSD，全称Single Shot MultiBox Detector，是Wei Liu在ECCV 2016上提出的一种目标检测算法，截至目前是主要的检测框架之一，相比Faster RCNN有明显的速度优势，相比YOLO又有明显的mAP优势（不过已经被CVPR 2017的YOLO9000超越）
在VOC2007上，SSD300比Faster R-CNN的FPS高了6.6倍

在VO...]]></description><category></category></item><item><title><![CDATA[卷积神经网络]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200520</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200520</guid><author>dihe3772</author><pubDate>Thu, 07 Mar 2019 19:33:00 +0800</pubDate><description><![CDATA[1.参数计算
卷积核大小*卷积核深度*输入该层数据的深度+卷积核深度
2.输出大小
(输入图像shape - 卷积核shape) / stride + 1
3.Dog-Vs-Cat实例
使用kaggle的Dog-Vs-Cat数据集，这个数据集包含25000条train数据，12500条test数据。这里仅使用train数据中的4000个数据作为全部数据。
数据集划分为2000...]]></description><category></category></item><item><title><![CDATA[boston_housing-多分类问题]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200538</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200538</guid><author>dihe3772</author><pubDate>Sun, 03 Mar 2019 20:22:00 +0800</pubDate><description><![CDATA[from keras.datasets import boston_housing
import numpy as np
from keras import models
from keras import layers
import matplotlib.pyplot as plt
#x,13个特征,一共404条数据
#y,连续值标签，单位是千美元
(x_train,...]]></description><category></category></item><item><title><![CDATA[reuters-多分类问题]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200537</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200537</guid><author>dihe3772</author><pubDate>Sun, 03 Mar 2019 14:03:00 +0800</pubDate><description><![CDATA[from keras.datasets import reuters
import numpy as np
from keras.utils.np_utils import to_categorical
from keras import layers
from keras import models
import matplotlib.pyplot as plt
def...]]></description><category></category></item><item><title><![CDATA[IMDB-二分类问题]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200524</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200524</guid><author>dihe3772</author><pubDate>Fri, 01 Mar 2019 16:32:00 +0800</pubDate><description><![CDATA[from keras.datasets import imdb
from keras.utils.np_utils import to_categorical
import numpy as np
from keras import models
from keras import layers
import matplotlib.pyplot as plt
#one-h...]]></description><category></category></item><item><title><![CDATA[神经网络学习技巧]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200533</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200533</guid><author>dihe3772</author><pubDate>Fri, 01 Mar 2019 10:06:00 +0800</pubDate><description><![CDATA[1.工作流程

定义问题

假设根据已知数据可以预测结果
面对的问题是什么类型问题，二分类、多分类、回归……


选择衡量模型好坏的指标，指标用来指示损失函数，常用的是准确率、召回率
数据分割，留出部分数据评估模型好坏（测试集），留出部分数据调节超参数（验证集），留出部分数据训练模型（训练集），验证集和训练集的分割可以使用留出验证集、K折交叉验证、重复k折交叉验证
选...]]></description><category></category></item><item><title><![CDATA[第一个神经网络]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200528</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200528</guid><author>dihe3772</author><pubDate>Wed, 27 Feb 2019 16:05:00 +0800</pubDate><description><![CDATA[1.keras训练神经网络的一般步骤

导入数据，做数据处理，使数据符合模型要求
定义网络结构
定义损失函数、优化器、监控指标
训练模型
图形化

2.使用MNIST数据集的一个例子

from keras.datasets import mnist
from keras import models
from keras import layers
from k...]]></description><category></category></item><item><title><![CDATA[朴素贝叶斯原理]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200531</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200531</guid><author>dihe3772</author><pubDate>Tue, 29 Jan 2019 20:29:00 +0800</pubDate><description><![CDATA[1.朴素贝叶斯的使用条件
朴素贝叶斯方法假设条件概率分布是相互独立的,这个假设会使每个属性独立的对分类结果产生影响
因此，根据独立性有，$p(x_1,x_2,...,x_n|C) = p(x_1|C)p(x_2|C)...p(x_n|C)，其中x_n是特征，C是分类的类别$
2.贝叶斯公式
由于朴素贝叶斯假设属性之间相互独立，所以贝叶斯公式可以写成如下形式
$p(C|\math...]]></description><category></category></item><item><title><![CDATA[集成方法 Bagging原理]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200529</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200529</guid><author>dihe3772</author><pubDate>Mon, 21 Jan 2019 16:24:00 +0800</pubDate><description><![CDATA[1.Bagging方法思路
Bagging独立的、并行的生成多个基本分类器，然后通过投票方式决定分类的类别
Bagging使用了自助法确定每个基本分类器的训练数据集，初始样本集中63.2%的数据会被采样到
从Training Sets中每次取1个，放回，再取1个，放回，重复直到取到n个组成Boot Strap1
同理生成Boot Strap2、Boot Strap3、……、Bo...]]></description><category></category></item><item><title><![CDATA[集成方法 Boosting原理]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200527</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200527</guid><author>dihe3772</author><pubDate>Fri, 28 Dec 2018 17:20:00 +0800</pubDate><description><![CDATA[1.Boosting方法思路
Boosting方法通过将一系列的基本分类器组合，生成更好的强学习器
基本分类器是通过迭代生成的，每一轮的迭代，会使误分类点的权重增大

Boosting方法常用的算法是AdaBoost(Adaptive Boosting)、GBDT(Gradient Boosting Decison Tree)
2.AdaBoost算法
算法要解决的2个问题...]]></description><category></category></item><item><title><![CDATA[scikit-learn 决策树 分类问题]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200525</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200525</guid><author>dihe3772</author><pubDate>Tue, 25 Dec 2018 10:19:00 +0800</pubDate><description><![CDATA[1.Demo

from sklearn import tree
import pydotplus
import numpy as np
#李航p59表数据
#年龄，有工作，有自己房子，信贷情况，类别
#青年0    中年1     老年2
#否0     是1
#一般0    好1      非常好2
datasets = np.array([['0', '0', ...]]></description><category></category></item><item><title><![CDATA[决策树原理]]></title><link>https://blog.youkuaiyun.com/dihe3772/article/details/101200523</link><guid>https://blog.youkuaiyun.com/dihe3772/article/details/101200523</guid><author>dihe3772</author><pubDate>Fri, 21 Dec 2018 16:53:00 +0800</pubDate><description><![CDATA[1.特征选择
特征选择是选取对训练数据有分类能力的特征，常用的有信息增益（information gain）、基尼不纯度（Gini impurity）
信息增益（information gain）
信息增益用在ID3、C4.5决策树生成算法中
wikipedia中信息增益的公式


J是分类数，几分类就是几
a是特征a的取值
Pr(i|a)是在a的属性下，第i...]]></description><category></category></item></channel></rss>