comparison caret_regression/tool3/predict.R @ 0:a4a2ad5a214e draft default tip

Uploaded
author deepakjadmin
date Thu, 05 Nov 2015 02:37:56 -0500
parents
children
comparison
equal deleted inserted replaced
-1:000000000000 0:a4a2ad5a214e
1 ##########
2 args <- commandArgs(T)
3 arg1 <- args[1]
4 arg2 <- args[2]
5 arg3 <- args[3]
6 #source("~/galaxy-dist/tools/mpdstoolsV2/tool3/Preold.R")
7 #pre(arg1,arg2,arg3)
8 set.seed(1)
9 pre <- function(args1,args2,args3){
10 #args <- commandArgs(TRUE)
11 nTrain <- read.csv(args1,row.names= 1, header = T) # example nTrain.csv file of unknown activity
12 #save(nTrain,file = "nTrain.RData")
13 #load("nTrain.RData")
14 load(args2) # model generated from previous programn
15 newdata <- nTrain
16 modelFit <- Fit
17 ###########
18 # input csv file must contaion the exact same column as used in model building #
19 # Also do pre-proccessing by means of centering and scaling
20 ## problem in s4 object so first check that the given model has s4 object in
21 ## >isS4(Fit$finalmodel) if it is s4 than add in with elseif loop
22 ## eg . isS4(plsFit$finalModel) == TRUE
23 f=function(x){
24 x<-as.numeric(as.character(x)) #first convert each column into numeric if it is from factor
25 x[is.na(x)] =median(x, na.rm=TRUE) #convert the item with NA to median value from the column
26 x #display the column
27 }
28
29 f2=function(x){
30 all(is.na(x))
31 }
32
33 fop <- apply(newdata,2,f2)
34 allcolumnmissing <- which(fop)
35 if (length(allcolumnmissing) > 0){
36 newdata[,allcolumnmissing] <- 0
37 newdata[,allcolumnmissing] <- newdata[,allcolumnmissing] + runif(3,0,0.000000000000000001) ### add noise}
38 }
39
40 library(caret)
41
42 #if(as.character(!isS4(Fit$finalModel == "TRUE")))
43 if((Fit$method != "svmRadial") && (Fit$method != "svmLinear") )
44 {
45 reqcol <- Fit$finalModel$xNames
46 newdata <- newdata[,reqcol]
47 newdata <- apply(newdata,2,f)
48 newdata <- newdata + runif(3,0,0.01) ### add noise to overcome from NZV error
49 newdata1 <- preProcess(newdata, method = c("center", "scale"))
50 newdata11 <- predict(newdata1,newdata)
51 ###########
52 library(stats)
53 testpredict <- predict(modelFit,newdata11)
54 names <- as.data.frame(rownames(nTrain))
55 colnames(names) <- "COMPOUND"
56 activity <- as.data.frame(testpredict)
57 colnames(activity) <- "PREDICTED VALUE"
58 dw <- format(cbind(names,activity),justify="centre")
59 write.table(dw,file=args3,row.names=FALSE,sep="\t")
60 }
61 #else if(Fit$method == "svmRadial")
62 else if((Fit$method == "svmLinear") | (Fit$method == "svmRadial"))
63 {
64 reqcol <- colnames(Fit$trainingData)
65 reqcol <- reqcol[1:length(reqcol)-1]
66 newdata <- newdata[,reqcol]
67 newdata <- apply(newdata,2,f)
68 newdata <- newdata + runif(3,0,0.01) ### add little noise to overcome from NZV problem
69 newdata1 <- preProcess(newdata, method = c("center", "scale"))
70 newdata11 <- predict(newdata1,newdata)
71 testpredict <- predict(modelFit,newdata11)
72 names <- as.data.frame(rownames(nTrain))
73 colnames(names) <- "COMPOUND"
74 activity <- as.data.frame(testpredict)
75 colnames(activity) <- "PREDICTED VALUE"
76 dw <- format(cbind(names,activity),justify="centre")
77 write.table(dw,file=args3,row.names=FALSE,sep="\t")
78
79 }
80 else {
81 dw <- "There is something wrong in data or model"
82 write.csv(dw,file=args3,row.names=FALSE)
83
84 }
85
86 }
87 pre(arg1,arg2,arg3)