0——1章
0、课程简介
建议先学习R语言有关的统计学课程
定义
构建方式
- 参数化模型和半参数化模型有回归系数beta;而机器学习没有参数。
- 正则化技术作为变量筛选方法：包括岭回归、lasso回归和弹性网络；而聚类分析和主成分与因子一般是降维分析（如：100多个变量聚类成10个变量）
image.png
课表
课表1
课表2
课表3
课表4
课表5
课表6
课表7
课表8
第一章 广义线性模型（generalize linear model，GLM）与R语言
image.png
image.png
image.png
image.png
image.png
**通过连接函数将非正态分布的函数变为正态分布的函数**
image.png
image.png
image.png
模型拟合和回归诊断
image.png
image.png
image.png
image.png
image.png
image.png
image.png
image.png
image.png
image.png
image.png
## 第1章代码开始
# 01）Generalized linear models 广义线性模型
# requires packages AER, robust, gcc
# install.packages(c("AER", "robust", "qcc"))
#####################################################################################################
## 02） Logistic Regression logistic回归
# get summary statistics
data(Affairs, package="AER")
#View(Affairs)
summary(Affairs)#展示结果
str(Affairs)#可以查看数据的结构
table(Affairs$affairs)
# create binary outcome variable
Affairs$ynaffair[Affairs$affairs > 0] <- 1
Affairs$ynaffair[Affairs$affairs == 0] <- 0
Affairs$ynaffair <- factor(Affairs$ynaffair,
levels=c(0,1),
labels=c("No","Yes"))
> table(Affairs$ynaffair)
No Yes
451 150
# fit full model
#注意此时的y变量是刚定义的ynaffair结局只有2类（有偷情的和无偷情的），可以二分类logit回归,因为用原始数据来构建模型只能用泊松回归，因为符合有限个数正整数结局时间。
fit.full <- glm(ynaffair ~ gender + age + yearsmarried + children +
religiousness + education + occupation +rating,
data=Affairs,family=binomial())#binomial就是logit回归
> summary(fit.full)#注意下面的带*的结果，都是有意义的。
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.37726 0.88776 1.551 0.120807 #Intercept表示常量
gendermale 0.28029 0.23909 1.172 0.241083
age -0.04426 0.01825 -2.425 0.015301 *
yearsmarried 0.09477 0.03221 2.942 0.003262 **
childrenyes 0.39767 0.29151 1.364 0.172508
religiousness -0.32472 0.08975 -3.618 0.000297 ***
education 0.02105 0.05051 0.417 0.676851
occupation 0.03092 0.07178 0.431 0.666630
rating -0.46845 0.09091 -5.153 2.56e-07 ***
---
Signif. codes:
0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 675.38 on 600 degrees of freedom
Residual deviance: 609.51 on 592 degrees of freedom
AIC: 627.51#AIC信息准则即Akaike information criterion，是衡量统计模型拟合优良性(Goodness of fit)的一种标准
# fit reduced model#将summary(fit.full)当中带*号的指标重新进行建模如下
fit.reduced <- glm(ynaffair ~ age + yearsmarried + religiousness +
rating, data=Affairs, family=binomial())
> summary(fit.reduced)
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.93083 0.61032 3.164 0.001558 **
age -0.03527 0.01736 -2.032 0.042127 *
yearsmarried 0.10062 0.02921 3.445 0.000571 ***
religiousness -0.32902 0.08945 -3.678 0.000235 ***
rating -0.46136 0.08884 -5.193 2.06e-07 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 675.38 on 600 degrees of freedom
Residual deviance: 615.36 on 596 degrees of freedom
AIC: 625.36
# compare models
> anova(fit.reduced, fit.full, test="Chisq")#比较两个模型的优劣，没有显著性
Analysis of Deviance Table
Model 1: ynaffair ~ age + yearsmarried + religiousness + rating
Model 2: ynaffair ~ gender + age + yearsmarried + children + religiousness +
education + occupation + rating
Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1 596 615.36
2 592 609.51 4 5.8474 0.2108
# interpret coefficients，
coef(fit.reduced)#查看模型的回归系数
exp(coef(fit.reduced))#OR值
exp(confint(fit.reduced)#OR值置信区间
# calculate probability of extramariatal affair by marital ratings#构建新的数据框，评分高低和婚外情的概率
testdata <- data.frame(rating = c(1, 2, 3, 4, 5),
age = mean(Affairs$age),
yearsmarried = mean(Affairs$yearsmarried),
religiousness = mean(Affairs$religiousness))
> testdata$prob <- predict(fit.reduced, newdata=testdata, type="response")
> testdata
rating age yearsmarried religiousness prob
1 1 32.48752 8.177696 3.116473 0.5302296
2 2 32.48752 8.177696 3.116473 0.4157377
3 3 32.48752 8.177696 3.116473 0.3096712
4 4 32.48752 8.177696 3.116473 0.2204547
5 5 32.48752 8.177696 3.116473 0.1513079
# calculate probabilites of extramariatal affair by age#年龄段和婚外情的概率
> testdata <- data.frame(rating = mean(Affairs$rating),
age = seq(17, 57, 10),
yearsmarried = mean(Affairs$yearsmarried),
religiousness = mean(Affairs$religiousness))
> testdata$prob <- predict(fit.reduced, newdata=testdata, type="response")
> testdata#17到26岁出轨概率更高
rating age yearsmarried religiousness prob
1 3.93178 17 8.177696 3.116473 0.3350834
2 3.93178 27 8.177696 3.116473 0.2615373
3 3.93178 37 8.177696 3.116473 0.1992953
4 3.93178 47 8.177696 3.116473 0.1488796
5 3.93178 57 8.177696 3.116473 0.1094738
# evaluate overdispersion#评价过度离势
> fit <- glm(ynaffair ~ age + yearsmarried + religiousness +
rating, family = binomial(), data = Affairs)#相当于logit回归
> fit.od <- glm(ynaffair ~ age + yearsmarried + religiousness +
rating, family = quasibinomial(), data = Affairs)#拟合过度离势；相当于稳健logit回归
> pchisq(summary(fit.od)$dispersion * fit$df.residual,
fit$df.residual, lower = F)#卡方检验的P值，大于0.05时，认为没有过度离势；如果有只要quasibinomial回归
[1] 0.340122
##########################################################################################
# Logistic回归案例2
Example11_4 <- read.table ("example11_4.csv", header=TRUE, sep=",")
attach(Example11_4)
fit1 <- glm(y~ x1 + x2, family= binomial(), data=Example11_4)
summary(fit1)
coefficients(fit1)
exp(coefficients(fit1))
exp (confint(fit1))
fit2 <- glm(y~ x1 + x2 + x1:x2 , family= binomial(), data=Example11_4)
summary(fit2)
coefficients(fit2)
exp(coefficients(fit2))
exp (confint(fit2))
Example11_4$x11 <- ifelse (x1==1 & x2==1, 1, 0)
Example11_4$x10 <- ifelse (x1==1 & x2==0, 1, 0)
Example11_4$x01 <- ifelse (x1==0 & x2==1, 1, 0)
Example11_4$x00 <- ifelse (x1==0 & x2==0, 1, 0)
fit3 <- glm(y~ x11 + x10 + x01, family= binomial(), data=Example11_4)
summary(fit3)
coefficients(fit3)
exp(coefficients(fit3))
exp(confint(fit3))
detach (Example11_4)
# Logistic回归案例3
Example11_5 <- read.table ("example11_5.csv", header=TRUE, sep=",")
attach(Example11_5)
fullfit <- glm(y~ x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 , family= binomial(), data=Example11_5)
summary(fullfit)
nothing <- glm(y~ 1, family= binomial(), data=Example11_5)
summary(nothing)
bothways <- step(nothing, list(lower=formula(nothing), upper=formula(fullfit)), direction="both")
fit1 <- glm(y~ x6 + x5 + x8 + x1 + x2 , family= binomial(), data=Example11_5)
summary(fit1)
fit2 <- glm(y~ x6 + x5 + x8 + x1, family= binomial(), data=Example11_5)
summary(fit2)
coefficients(fit2)
exp(coefficients(fit2))
exp (confint(fit2))
detach (Example11_5)
#条件Logistic回归案例 4
#install.packages("survival")
library(survival)
Example11_6 <- read.table ("example11_6.csv", header=TRUE, sep=",")
attach(Example11_6)
model <- clogit(outcome~ exposure+ strata(id))
summary(model)
detach(Example11_6)
###################################################################################################
# 03）列线图
#Logistic回归案例 5 低出生体重儿列线图
library(foreign)
library(rms)
mydata<-read.spss("lweight.sav")
mydata<-as.data.frame(mydata)
head(mydata)
mydata$low <- ifelse(mydata$low =="低出生体重",1,0)
mydata$race1 <- ifelse(mydata$race =="白种人",1,0)
mydata$race2 <- ifelse(mydata$race =="黑种人",1,0)
mydata$race3 <- ifelse(mydata$race =="其他种族",1,0)
attach(mydata)
dd<-datadist(mydata)
options(datadist='dd')
fit1<-lrm(low~age+ftv+ht+lwt+ptl+smoke+ui+race1+race2,data=mydata,x=T,y=T)
fit1
summary(fit1)
nom1 <- nomogram(fit1, fun=plogis,fun.at=c(.001, .01, .05, seq(.1,.9, by=.1), .95, .99, .999),lp=F, funlabel="Low weight rate")
plot(nom1)
cal1 <- calibrate(fit1, method='boot', B=1000)
plot(cal1,xlim=c(0,1.0),ylim=c(0,1.0))
mydata$race <- as.factor(ifelse(mydata$race=="白种人", "白种人","黑人及其他种族"))
dd<-datadist(mydata)
options(datadist='dd')
fit2<-lrm(low~age+ftv+ht+lwt+ptl+smoke+ui+race,data=mydata,x=T,y=T)
fit2
summary(fit2)
nom2 <- nomogram(fit2, fun=plogis,fun.at=c(.001, .01, .05, seq(.1,.9, by=.1), .95, .99, .999),lp=F, funlabel="Low weight rate")
plot(nom2)
cal2 <- calibrate(fit2, method='boot', B=1000)
plot(cal2,xlim=c(0,1.0),ylim=c(0,1.0))
fit3<-lrm(low~ht+lwt+ptl+smoke+race,data=mydata,x=T,y=T)
fit3
summary(fit3)
nom3 <- nomogram(fit3, fun=plogis,fun.at=c(.001, .01, .05, seq(.1,.9, by=.1), .95, .99, .999),lp=F, funlabel="Low weight rate")
plot(nom3)
cal3 <- calibrate(fit3, method='boot', B=1000)
plot(cal3,xlim=c(0,1.0),ylim=c(0,1.0))
#########################################################################################################
# 04）C-statistics计算
library(foreign)
library(rms)
mydata<-read.spss("lweight.sav")
mydata<-as.data.frame(mydata)
head(mydata)
mydata$low <- ifelse(mydata$low =="低出生体重",1,0)
mydata$race1 <- ifelse(mydata$race =="白种人",1,0)
mydata$race2 <- ifelse(mydata$race =="黑种人",1,0)
mydata$race3 <- ifelse(mydata$race =="其他种族",1,0)
attach(mydata)
dd<-datadist(mydata)
options(datadist='dd')
fit1<-lrm(low~age+ftv+ht+lwt+ptl+smoke+ui+race1+race2,data=mydata,x=T,y=T)
fit1 #直接读取模型中Rank Discrim.参数 C
mydata$predvalue<-predict(fit1)
library(ROCR)
pred <- prediction(mydata$predvalue, mydata$low)
perf<- performance(pred,"tpr","fpr")
plot(perf)
abline(0,1)
auc <- performance(pred,"auc")
auc #auc即是C-statistics
#library(Hmisc)
somers2(mydata$predvalue, mydata$low) #somers2 {Hmisc}
######################################################################################################
## 05）亚组分析森林图
#Logistic回归案例 6 亚组分析森林图
library(forestplot)
rs_forest <- read.csv('rs_forest.csv',header = FALSE)
# 读入数据的时候大家一定要把header设置成FALSE，保证第一行不被当作列名称。
# tiff('Figure 1.tiff',height = 1600,width = 2400,res= 300)
forestplot(labeltext = as.matrix(rs_forest[,1:3]),
#设置用于文本展示的列，此处我们用数据的前三列作为文本，在图中展示
mean = rs_forest$V4, #设置均值
lower = rs_forest$V5, #设置均值的lowlimits限
upper = rs_forest$V6, #设置均值的uplimits限
is.summary = c(T,T,T,F,F,T,F,F,T,F,F),
#该参数接受一个逻辑向量，用于定义数据中的每一行是否是汇总值，若是，则在对应位置设置为TRUE，若否，则设置为FALSE；设置为TRUE的行则以粗体出现
zero = 1, #设置参照值，此处我们展示的是OR值，故参照值是1，而不是0
boxsize = 0.4, #设置点估计的方形大小
lineheight = unit(10,'mm'),#设置图形中的行距
colgap = unit(3,'mm'),#设置图形中的列间距
lwd.zero = 2,#设置参考线的粗细
lwd.ci = 1.5,#设置区间估计线的粗细
col=fpColors(box='#458B00', summary= "#8B008B",lines = 'black',zero = '#7AC5CD'),
#使用fpColors()函数定义图形元素的颜色，从左至右分别对应点估计方形，汇总值，区间估计线，参考线
xlab="The estimates",#设置x轴标签
graph.pos = 3)#设置森林图的位置，此处设置为3，则出现在第三列
##################################################################################################################
## 06） Poisson Regression
# look at dataset
data(breslow.dat, package="robust")
names(breslow.dat)
summary(breslow.dat[c(6, 7, 8, 10)])
str(breslow.dat[c(6, 7, 8, 10)])
# plot distribution of post-treatment seizure counts
opar <- par(no.readonly=TRUE)
par(mfrow=c(1, 2))
attach(breslow.dat)
hist(sumY, breaks=20, xlab="Seizure Count",
main="Distribution of Seizures")
boxplot(sumY ~ Trt, xlab="Treatment", main="Group Comparisons")
par(opar)
# fit regression#拟合回归
> fit <- glm(sumY ~ Base + Age + Trt, data=breslow.dat, family=poisson())
> summary(fit)
Call:
glm(formula = sumY ~ Base + Age + Trt, family = poisson(), data = breslow.dat)
Deviance Residuals:
Min 1Q Median 3Q Max
-6.0569 -2.0433 -0.9397 0.7929 11.0061
#哪些因素是影响癫痫发作的，看Coefficients:
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.9488259 0.1356191 14.370 < 2e-16 ***#截距是常量
Base 0.0226517 0.0005093 44.476 < 2e-16 ***#影响
Age 0.0227401 0.0040240 5.651 1.59e-08 ***#影响
Trtprogabide -0.1527009 0.0478051 -3.194 0.0014 ** #影响
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for poisson family taken to be 1)
Null deviance: 2122.73 on 58 degrees of freedom
Residual deviance: 559.44 on 55 degrees of freedom
AIC: 850.71
Number of Fisher Scoring iterations: 5
# interpret model parameters
> coef(fit)
(Intercept) Base Age Trtprogabide
1.94882593 0.02265174 0.02274013 -0.15270095
> exp(coef(fit))#接下来，看回归系数的反对数，泊松回归不太容易解释
(Intercept) Base Age Trtprogabide
7.0204403 1.0229102 1.0230007 0.8583864
###泊松回归不像logit回归不太容易解释exp(coef(fit))中Trtprogabide 药物和对照组的比0.8583864 的关系
# evaluate overdispersion
> deviance(fit)/df.residual(fit)
> library(qcc)
> qcc.overdispersion.test(breslow.dat$sumY, type="poisson")
Overdispersion test Obs.Var/Theor.Var Statistic p-value
poisson data 62.87013 3646.468 0
##这里的0表示有有过度离势，只能用稳健的泊松分布，发现只有一个有意义的。
# fit model with quasipoisson#用quasipoisson评估过度离势是泊松分布的特点，根据之前评价过度离势可以进行两模型的比较，这里结果三个因素都是有统计学意义的因素，所以这里考虑用qcc的包，qcc.overdispersion.test检验
fit.od <- glm(sumY ~ Base + Age + Trt, data=breslow.dat,
family=quasipoisson())
summary(fit.od)
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1.948826 0.465091 4.190 0.000102 ***
Base 0.022652 0.001747 12.969 < 2e-16 ***###只有这有关系
Age 0.022740 0.013800 1.648 0.105085
Trtprogabide -0.152701 0.163943 -0.931 0.355702 ###否定了药物的作用，统计和临床有差别
## 第1章代码结束
泊松分布案例
image.png
- logistic回归就是指“有”和“无”
- 泊松分布指的是非0的整数
网友评论