[Lecture Notes]Genevera Allen:Statistical Learning using R and Matlab

1. ######################
   
2. #lecture 24 - boosting & bagging
   
3. 
   
4. require(rpart)
   
5. require(randomForest)
   
6. 
   
7. 
   
8. #reading in data
   
9. data = read.csv("spam_dat.csv",header=FALSE)
   
10. ndat = read.delim("spam_vars.txt",header=FALSE)
    
11. 
    
12. #parsing variable names
    
13. nams = NULL
    
14. for(i in 1:nrow(ndat))
    
15. {
    
16.   vec = strsplit(as.character(ndat[i,]),split="_")
    
17.   for(j in 1:length(vec[[1]]))
    
18.     {
    
19.       if(length(grep(":",vec[[1]][j]))>0)
    
20.         {
    
21.           vars = strsplit(vec[[1]][j],split=":")
    
22.           nams = c(nams,vars[[1]][1])
    
23.         }
    
24.     }
    
25. }
    
26. 
    
27. Y = data[,58]
    
28. n = length(Y)
    
29. sum(Y)/n
    
30. X = as.matrix(log(1 + data[,1:57]))
    
31. colnames(X) = nams
    
32. X = scale(X)/sqrt(n-1)
    
33. ind = sample(1:nrow(X),floor(nrow(X)*.3))
    
34. xts = X[ind,]; xtr = X[-ind,]
    
35. yts = Y[ind]; ytr = Y[-ind]
    
36. trfdat = data.frame(as.factor(ytr),xtr)
    
37. names(trfdat)[1] = "ytr"
    
38. trdat = data.frame(ytr,xtr)
    
39. tsdat = data.frame(yts,xts)
    
40. 
    
41. 
    
42. #trees
    
43. fit1 = rpart(ytr~.,data=trdat,method="class",xval=5,cp=.01)
    
44. trpred = apply(predict(fit1),1,which.max) - 1
    
45. tree.trerr = sum(abs(ytr - trpred))/length(ytr)
    
46. tspred = apply(predict(fit1,newdata=data.frame(xts)),1,which.max) - 1
    
47. tree.tserr = sum(abs(yts - tspred))/length(yts)
    
48. tree.err = c(tree.trerr, tree.tserr)
    
49. tree.err
    
50. 
    
51. 
    
52. #logistic regression
    
53. fit2 = glm(ytr~.,data=trdat,family="binomial")
    
54. trpred = (sign(predict(fit2)) + 1)/2
    
55. logit.trerr = sum(abs(ytr - trpred))/length(ytr)
    
56. tspred = (sign(predict(fit2,newdata=data.frame(xts))) + 1)/2
    
57. logit.tserr = sum(abs(yts - tspred))/length(yts)
    
58. logit.err = c(logit.trerr, logit.tserr)
    
59. logit.err
    
60. 
    
61. 
    
62. #######boosting
    
63. require(ada)
    
64. 
    
65. #adaboost
    
66. fit3 = ada(x=xtr,y=ytr,loss="exponential",type="discrete",iter=150)
    
67. ab.trerr = (fit3$confusion[2,1] + fit3$confusion[1,2])/length(ytr)
    
68. plot(fit3)
    
69. tspred = round(predict(fit3,newdata=data.frame(xts),type="probs"))
    
70. ab.tserr = sum(abs(yts - (apply(tspred,1,which.max)-1)))/length(yts)
    
71. ab.err = c(ab.trerr, ab.tserr)
    
72. ab.err
    
73. 
    
74. #logitboost
    
75. fit4 = ada(x=xtr,y=ytr,loss="logistic",type="real",iter=150)
    
76. lb.trerr = (fit4$confusion[2,1] + fit4$confusion[1,2])/length(ytr)
    
77. plot(fit4)
    
78. tspred = round(predict(fit4,newdata=data.frame(xts),type="probs"))
    
79. lb.tserr = sum(abs(yts - (apply(tspred,1,which.max)-1)))/length(yts)
    
80. lb.err = c(lb.trerr, lb.tserr)
    
81. lb.err
    
82. 
    
83. 
    
84. 
    
85. #bagging
    
86. fit5 = randomForest(ytr~.,mtry=57,data=trfdat,xtest=data.frame(xts),ytest=as.factor(yts))
    
87. print(fit5)
    
88. bag.err = c(sum(abs(ytr-(as.numeric(fit5$predicted)-1))/length(ytr)),(fit5$err.rate[500,2]*(length(yts)-sum(yts))+fit5$err.rate[500,3]*sum(yts))/length(yts))
    
89. bag.err
    
90. 
    
91. 
    
92. errmat = cbind(tree.err,logit.err,ab.err,lb.err,bag.err)
    
93. errmat

复制代码

1. ######################
   
2. #lecture 25 - Random Forest, boosting & bagging
   
3. 
   
4. require(rpart)
   
5. require(randomForest)
   
6. require(ada)
   
7. 
   
8. #reading in data
   
9. data = read.csv("spam_dat.csv",header=FALSE)
   
10. ndat = read.delim("spam_vars.txt",header=FALSE)
    
11. 
    
12. #parsing variable names
    
13. nams = NULL
    
14. for(i in 1:nrow(ndat))
    
15. {
    
16.   vec = strsplit(as.character(ndat[i,]),split="_")
    
17.   for(j in 1:length(vec[[1]]))
    
18.     {
    
19.       if(length(grep(":",vec[[1]][j]))>0)
    
20.         {
    
21.           vars = strsplit(vec[[1]][j],split=":")
    
22.           nams = c(nams,vars[[1]][1])
    
23.         }
    
24.     }
    
25. }
    
26. 
    
27. Y = data[,58]
    
28. n = length(Y)
    
29. sum(Y)/n
    
30. X = as.matrix(log(1 + data[,1:57]))
    
31. colnames(X) = nams
    
32. X = scale(X)/sqrt(n-1)
    
33. ind = sample(1:nrow(X),floor(nrow(X)*.3))
    
34. xts = X[ind,]; xtr = X[-ind,]
    
35. yts = Y[ind]; ytr = Y[-ind]
    
36. trfdat = data.frame(as.factor(ytr),xtr)
    
37. names(trfdat)[1] = "ytr"
    
38. trdat = data.frame(ytr,xtr)
    
39. tsdat = data.frame(yts,xts)
    
40. 
    
41. 
    
42. #trees
    
43. fit1 = rpart(ytr~.,data=trdat,method="class",xval=5,cp=.01)
    
44. trpred = apply(predict(fit1),1,which.max) - 1
    
45. tree.trerr = sum(abs(ytr - trpred))/length(ytr)
    
46. tspred = apply(predict(fit1,newdata=data.frame(xts)),1,which.max) - 1
    
47. tree.tserr = sum(abs(yts - tspred))/length(yts)
    
48. tree.err = c(tree.trerr, tree.tserr)
    
49. tree.err
    
50. 
    
51. 
    
52. #logistic regression
    
53. fit2 = glm(ytr~.,data=trdat,family="binomial")
    
54. trpred = (sign(predict(fit2)) + 1)/2
    
55. logit.trerr = sum(abs(ytr - trpred))/length(ytr)
    
56. tspred = (sign(predict(fit2,newdata=data.frame(xts))) + 1)/2
    
57. logit.tserr = sum(abs(yts - tspred))/length(yts)
    
58. logit.err = c(logit.trerr, logit.tserr)
    
59. logit.err
    
60. 
    
61. 
    
62. #######boosting
    
63. 
    
64. #adaboost
    
65. fit3 = ada(x=xtr,y=ytr,loss="exponential",type="discrete",iter=100)
    
66. ab.trerr = (fit3$confusion[2,1] + fit3$confusion[1,2])/length(ytr)
    
67. plot(fit3)
    
68. tspred = round(predict(fit3,newdata=data.frame(xts),type="probs"))
    
69. ab.tserr = sum(abs(yts - (apply(tspred,1,which.max)-1)))/length(yts)
    
70. ab.err = c(ab.trerr, ab.tserr)
    
71. ab.err
    
72. 
    
73. #logitboost
    
74. fit4 = ada(x=xtr,y=ytr,loss="logistic",type="real",iter=100)
    
75. lb.trerr = (fit4$confusion[2,1] + fit4$confusion[1,2])/length(ytr)
    
76. plot(fit4)
    
77. tspred = round(predict(fit4,newdata=data.frame(xts),type="probs"))
    
78. lb.tserr = sum(abs(yts - (apply(tspred,1,which.max)-1)))/length(yts)
    
79. lb.err = c(lb.trerr, lb.tserr)
    
80. lb.err
    
81. 
    
82. errmat = cbind(tree.err,logit.err,ab.err,lb.err)
    
83. errmat
    
84. 
    
85. 
    
86. #bagging
    
87. fit5 = randomForest(ytr~.,mtry=57,data=trfdat,xtest=data.frame(xts),ytest=as.factor(yts))
    
88. print(fit5)
    
89. bag.err = c(sum(abs(ytr-(as.numeric(fit5$predicted)-1))/length(ytr)),(fit5$err.rate[500,2]*(length(yts)-sum(yts))+fit5$err.rate[500,3]*sum(yts))/length(yts))
    
90. bag.err
    
91. 
    
92. 
    
93. #random forest
    
94. fit6 = randomForest(ytr~.,mtry=4,data=trfdat,xtest=data.frame(xts),ytest=as.factor(yts))
    
95. print(fit6)
    
96. rf.err = c(sum(abs(ytr-(as.numeric(fit6$predicted)-1))/length(ytr)),(fit6$err.rate[500,2]*(length(yts)-sum(yts))+fit6$err.rate[500,3]*sum(yts))/length(yts))
    
97. rf.err
    
98. 
    
99. 
    
100. errmat = cbind(tree.err,logit.err,ab.err,lb.err,bag.err,rf.err)
     
101. errmat

复制代码

感谢楼主分享，辛苦了。

ding!!!!!!!!!

Genevera Allen:Statistical Learning using R and Matlab

感谢楼主分享