2022/10/31 10:55 R Notebook

R Notebook
Code

Hide

library(class)
library(ggplot2)
library(dplyr)

Attaching package: ‘dplyr’

The following objects are masked from ‘package:stats’:

filter, lag

The following objects are masked from ‘package:base’:

intersect, setdiff, setequal, union

Hide

library(glmnet)

Loading required package: Matrix

Loaded glmnet 4.1-4

Hide

library(Hmisc)

file:///Users/junkong/Desktop/1.nb.html 1/11
2022/10/31 10:55 R Notebook

Loading required package: lattice

Loading required package: survival
Loading required package: Formula
Registered S3 methods overwritten by 'htmltools':
method from
print.html tools:rstudio
print.shiny.tag tools:rstudio
print.shiny.tag.list tools:rstudio
Registered S3 method overwritten by 'htmlwidgets':
method from
print.htmlwidget tools:rstudio
Registered S3 method overwritten by 'data.table':
method from
print.data.table

Attaching package: ‘Hmisc’

The following objects are masked from ‘package:dplyr’:

src, summarize

The following objects are masked from ‘package:base’:

format.pval, units

Load the data

Hide

file:///Users/junkong/Desktop/1.nb.html 2/11
2022/10/31 10:55 R Notebook

df = read.csv("/Users/junkong/Desktop/ranial/sensor.csv")
df$sensor_00=impute(df$sensor_00,median)
df$sensor_01=impute(df$sensor_01,median)
df$sensor_02=impute(df$sensor_02,median)
df$sensor_03=impute(df$sensor_03,median)
df$sensor_04=impute(df$sensor_04,median)
df$sensor_05=impute(df$sensor_05,median)
df$sensor_06=impute(df$sensor_06,median)
df$sensor_07=impute(df$sensor_07,median)
df$sensor_08=impute(df$sensor_08,median)
df$sensor_09=impute(df$sensor_09,median)
df$sensor_10=impute(df$sensor_10,median)

df$sensor_11=impute(df$sensor_11,median)
df$sensor_12=impute(df$sensor_12,median)
df$sensor_13=impute(df$sensor_13,median)
df$sensor_14=impute(df$sensor_14,median)
df$sensor_15=impute(df$sensor_15,median)
df$sensor_16=impute(df$sensor_16,median)
df$sensor_17=impute(df$sensor_17,median)
df$sensor_18=impute(df$sensor_18,median)
df$sensor_19=impute(df$sensor_19,median)
df$sensor_20=impute(df$sensor_20,median)

df$sensor_21=impute(df$sensor_21,median)
df$sensor_22=impute(df$sensor_22,median)
df$sensor_23=impute(df$sensor_23,median)
df$sensor_24=impute(df$sensor_24,median)
df$sensor_25=impute(df$sensor_25,median)
df$sensor_26=impute(df$sensor_26,median)
df$sensor_27=impute(df$sensor_27,median)
df$sensor_28=impute(df$sensor_28,median)
df$sensor_29=impute(df$sensor_29,median)
df$sensor_30=impute(df$sensor_30,median)

df$sensor_31=impute(df$sensor_31,median)
df$sensor_32=impute(df$sensor_32,median)
df$sensor_33=impute(df$sensor_33,median)
df$sensor_34=impute(df$sensor_34,median)
df$sensor_35=impute(df$sensor_35,median)
df$sensor_36=impute(df$sensor_36,median)
df$sensor_37=impute(df$sensor_37,median)
df$sensor_38=impute(df$sensor_38,median)
df$sensor_39=impute(df$sensor_39,median)
df$sensor_40=impute(df$sensor_40,median)

df$sensor_41=impute(df$sensor_41,median)
df$sensor_42=impute(df$sensor_42,median)
df$sensor_43=impute(df$sensor_43,median)
df$sensor_44=impute(df$sensor_44,median)
df$sensor_45=impute(df$sensor_45,median)
df$sensor_46=impute(df$sensor_46,median)
df$sensor_47=impute(df$sensor_47,median)
df$sensor_48=impute(df$sensor_48,median)
df$sensor_49=impute(df$sensor_49,median)

file:///Users/junkong/Desktop/1.nb.html 3/11
2022/10/31 10:55 R Notebook
df$sensor_50=impute(df$sensor_50,median)
df$sensor_51=impute(df$sensor_51,median)

df$machine_status[df$machine_status == "NORMAL"] <- 0

df$machine_status[df$machine_status == "BROKEN"] <- 1
df$machine_status[df$machine_status == "RECOVERING"] <- 1

num_train <- nrow(df) * 0.7

inTrain <- sample(nrow(df), size = num_train)

train <- df[inTrain,]

test <- df[-inTrain,]

X= train[,3:54]
X.train = X[,c(1:15,17:50,52)]
X.train = data.matrix(X.train)
y.train<- as.numeric(matrix(c(train[,55]),byrow = T))

X_ = test[,3:54]
X.test =X_[,c(1:15,17:50,52)]
X.test = data.matrix(X.test)
y.test = as.numeric(matrix(c(test[,55]),byrow = T))

Logistic Regression
Hide

# model 1: ridge alpha = 0

cv.fit.1 = cv.glmnet(X.train, y.train, alpha = 0, family = "binomial" )

plot(cv.fit.1)

file:///Users/junkong/Desktop/1.nb.html 4/11
2022/10/31 10:55 R Notebook

Hide

fit1 = glmnet(X.train, y.train, alpha = 0, family = "binomial")

plot(fit1)

Hide

file:///Users/junkong/Desktop/1.nb.html 5/11
2022/10/31 10:55 R Notebook

fit1 = glmnet(X.train, y.train, alpha = 0, lambda = cv.fit.1$lambda.min,

family = "binomial")
y.train.hat1 = predict(fit1, newx = X.train, type = "response")
y.test.hat1 = predict(fit1, newx = X.test, type = "response")
train.err1 = mean((y.train.hat1 - y.train)^2)
test.err1 = mean((y.test.hat1 - y.test)^2)
paste(c("ridge train mse is", "ridge test mse is"), round(c(train.err1, test.err1), 6
))

[1] "ridge train mse is 0.066066" "ridge test mse is 0.064981"

Hide

barplot(as.vector(fit1$beta), main = paste("alpha = ", 0))

Hide

# model 2: elastic-netnet alpha = 0.5

cv.fit.2 = cv.glmnet(X.train, y.train, alpha = 0.5, family = "binomial" )
plot(cv.fit.2)

file:///Users/junkong/Desktop/1.nb.html 6/11
2022/10/31 10:55 R Notebook

Hide

fit2 = glmnet(X.train, y.train, alpha = 0.5, family = "binomial")

plot(fit2)

Hide

file:///Users/junkong/Desktop/1.nb.html 7/11
2022/10/31 10:55 R Notebook

fit2 = glmnet(X.train, y.train, alpha = 0.5, lambda = cv.fit.2$lambda.mi

n, family = "binomial")
y.train.hat2 = predict(fit2, newx = X.train, type = "response")
y.test.hat2 = predict(fit2, newx = X.test, type = "response")
train.err2 = mean((y.train.hat2 - y.train)^2)
test.err2 = mean((y.test.hat2 - y.test)^2)
paste(c("el-net train mse is", "el-net test mse is"), round(c(train.err2, test.err2),
6))

[1] "el-net train mse is 0.931859" "el-net test mse is 0.932825"

Hide

barplot(as.vector(fit2$beta), main = paste("alpha = ", 0.5))

Hide

# model 3: lasso alpha = 1

cv.fit.3 = cv.glmnet(X.train, y.train, alpha = 1, family = "binomial" )
plot(cv.fit.3)

file:///Users/junkong/Desktop/1.nb.html 8/11
2022/10/31 10:55 R Notebook

Hide

fit3 = glmnet(X.train, y.train, alpha = 1, family = "binomial")

plot(fit3)

Hide

file:///Users/junkong/Desktop/1.nb.html 9/11
2022/10/31 10:55 R Notebook

fit3 = glmnet(X.train, y.train, alpha = 1, lambda = cv.fit.3$lambda.min,

family = "binomial")

Warning: from glmnet C++ code (error code -1); Convergence for 1th lambda value not r
eached after maxit=100000 iterations; solutions for larger lambdas returnedWarning: a
n empty model has been returned; probably a convergence issue

Hide

y.train.hat3 = predict(fit3, newx = X.train, type = "response")

y.test.hat3 = predict(fit3, newx = X.test, type = "response")
train.err3 = mean((y.train.hat3 - y.train)^2)
test.err3 = mean((y.test.hat3 - y.test)^2)
paste(c("lasso train mse is", "lasso test mse is"), round(c(train.err3, test.err3), 6
))

[1] "lasso train mse is 0.25" "lasso test mse is 0.25"

Hide

barplot(as.vector(fit3$beta), main = paste("alpha = ", 1))

KNN
Hide

file:///Users/junkong/Desktop/1.nb.html 10/11
2022/10/31 10:55 R Notebook

k.values = seq(2,20, by = 1) # in KNN, these are the values of k

k.length = length(k.values)
test.err = matrix(0, k.length)
train.err = matrix(0, k.length)
i = 1

n.train = dim(train)[1]
n.test = dim(test)[1]
n = n.test + n.train

for (ki in k.values){

y.test.hat = knn(X.train, X.test, y.train, ki)
test.err[i] = sum(y.test.hat != y.test)/n.test
y.train.hat = knn(X.train, X.train, y.train, ki)
train.err[i] = sum(y.train.hat != y.train)/n.train
print(paste("ki=",ki, ", test.err=", test.err[i],", train.err=", train.err[i]))
i=i+1
}

[1] "ki= 2 , test.err= 0.000393367223432583 , train.err= 0.0001491337275651"

[1] "ki= 3 , test.err= 0.000378237714839022 , train.err= 0.000142649652453574"
[1] "ki= 4 , test.err= 0.000378237714839022 , train.err= 0.000201006328457309"

file:///Users/junkong/Desktop/1.nb.html 11/11