-
Notifications
You must be signed in to change notification settings - Fork 0
/
Model tester on Generated data .R
305 lines (254 loc) · 17.2 KB
/
Model tester on Generated data .R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
#########QUICK IS MORE UP TO DATE!!!!#######
########DO NOT USE UNTIL MERGING!!!!!!#######
## capture messages and errors to a file.
zz <- file("all.Rout", open="wt")
sink(zz, type="message")
try(log("a"))
list.of.packages <- c("caret","caretEnsemble","mlr","MLmetrics")
#list.of.packages <- c("caretEnsemble","logicFS"," RWeka","ordinalNet","xgboost","mlr","caret","MLmetrics","bartMachine","spikeslab","party","rqPen","monomvn","foba","logicFS","rPython","qrnn","randomGLM","msaenet","Rborist","relaxo","ordinalNet","rrf","frbs","extraTrees","ipred","elasticnet","bst","brnn","Boruta","arm","elmNN","evtree","extraTrees","deepnet","kknn","KRLS","RSNNS","partDSA","plsRglm","quantregForest","ranger","inTrees")
new.packages <- list.of.packages[!(list.of.packages %in% installed.packages()[,"Package"])]
if(length(new.packages)) install.packages(new.packages, dep = TRUE)
#install.packages("mlr", dependencies = c("Depends", "Suggests"))
#install.packages("caret", dependencies = c("Depends", "Suggests"))
#install.packages("SuperLearner", dependencies = c("Depends", "Suggests"))
#install.packages("rattle", dependencies = c("Depends", "Suggests"))
# Load libraries
#library(mlbench)
library(caret)
#library(caretEnsemble)
library(MLmetrics)
before.last.alg<-as.matrix(read.csv("beforelast algorithm.csv", sep = ",",fill=TRUE, header = FALSE,quote="",dec="."))
last.alg<-as.matrix(read.csv("last algorithm tried.csv", sep = ",",fill=TRUE, header = FALSE,quote="",dec="."))
if(last.alg==before.last.alg){print("algorithm may be broken")}
write.table(last.alg,file = "beforelast algorithm.csv", quote = F, row.names = F,col.names = F)
#make sure not to redo a test
check.redundant<-function(df=df.previous.calcs,norming="asis",trans.y=1,withextra="missing",missingdata="leaveempty",datasource="mean" ,column.to.predict=200,allmodel="ctree")
{
for(intern in 1:length(df[,1])){
if((any(df[intern,] == norming, na.rm=T))&&
(any(df[intern,] == withextra, na.rm=T))&&
(any(df[intern,] == missingdata, na.rm=T))&&
(any(df[intern,] == datasource, na.rm=T))&&
(any(df[intern,] == column.to.predict, na.rm=T))&&
(any(df[intern,] == allmodel, na.rm=T))&&
( (df[intern,9] == trans.y)))
{return(TRUE)}
}
return(FALSE)
}
allmodels <- c("avNNet", "bagEarth", "bagEarthGCV",
"bayesglm", "bdk", "blackboost", "Boruta", "brnn", "BstLm" ,
"bstTree", "cforest", "ctree", "ctree2", "cubist", "DENFIS",
"dnn", "earth", "elm", "enet", "evtree",
"extraTrees", "gamLoess", "gaussprLinear", "gaussprPoly", "gaussprRadial",
"gcvEarth","glm", "glmboost", "icr", "kernelpls",
"kknn", "knn", "krlsRadial", "lars" , "lasso",
"leapBackward", "leapForward", "leapSeq", "lm", "M5", "M5Rules",
"mlpWeightDecay", "neuralnet" , "partDSA",
"pcaNNet", "pcr", "penalized", "pls", "plsRglm", "ppr",
"qrf" , "ranger", "rf")
allmodels <- c("rlm", "rpart", "rpart2",
"RRF", "RRFglobal", "simpls",
"svmLinear", "svmPoly", "svmRadial", "svmRadialCost",
"widekernelpls", "xgbLinear",
"xgbTree")
allmodels <- c("avNNet","BstLm","bstTree","cforest","ctree","ctree2",
"cubist","earth","enet","evtree","glmboost",
"icr","kernelpls","kknn","lasso","pcaNNet",
"pcr","pls","qrf","ranger","rf"
)
#allmodels <- c("ranger")#"BstLm","enet","lasso",
#allmodels <- c("rf")"rqlasso",, "xyf" "rvmPoly", "rvmRadial", "spls", "superpc" , "treebag", "svmLinear2", "SBC",
#allmodels <- c("bartMachine", "xgbLinear", "pcaNNet","svmLinear","glmnet","cforest","cubist","rf","ranger")"glmnet",
#wow rfRules is really slow "rfRules","WM", takes 50min
# brak everythig "rbfDDA","ridge","rqnc",
# use "rf" to test all
allmodels <- unique(modelLookup()[modelLookup()$forReg,c(1)])
#library(doParallel); cl <- makeCluster(detectCores()); registerDoParallel(cl)
adaptControl <- trainControl(method = "adaptive_cv",
number = 8, repeats = 5,
adaptive = list(min = 4, alpha = 0.05,
method = "gls", complete = FALSE),
search = "random")
simpleControl <- trainControl(method = "cv",
number = 10,
search = "random")
seed.const=222
seed.var=seed.const
column.to.predict=1
print(date());
if(!exists("gen.count")){gen.count=17}
gens.names<-as.matrix(read.table("gens names.csv", sep = ",",header = FALSE,row.names=1,fill=TRUE, quote="",dec="."))
for(gend.data in 13:gen.count){
data.source<-as.matrix(read.csv(paste(gens.names[gend.data],".csv", sep = ""), sep = ",",fill=TRUE, header = FALSE,quote="",dec="."))
datasource<-gens.names[gend.data]
missingdatas=c("ignore")
for(missingdata in missingdatas){
withextras=c("none")
for(withextra in withextras){
################data wrestling###############
dependant.selection=complete.cases(data.source[,column.to.predict])
df.previous.calcs=as.data.frame(read.csv(file="gen test out.csv", header = FALSE, sep = ",", quote = "",
dec = ".", fill = TRUE, comment.char = ""))
unimportant.computations<-vector(mode = "logical",length=length(df.previous.calcs[,1]) )
for(intern in 1:length(df.previous.calcs[,1])){
if((any(df.previous.calcs[intern,] == withextra, na.rm=T))&&
(any(df.previous.calcs[intern,] == missingdata, na.rm=T))&&
(any(df.previous.calcs[intern,] == datasource, na.rm=T))&&
(any(df.previous.calcs[intern,] == column.to.predict, na.rm=T)))
{unimportant.computations[intern]<-T}}
df.previous.calcs<-df.previous.calcs[unimportant.computations,]
#data.source=data.frame( data.source[,column.to.predict],data.source[,1:2], data.source[,4:(column.to.predict-1)], data.source[,(column.to.predict+1):length( data.source[1,])])
normings=c("range01")#,"asis","expoTrans","quantile","centernscale","YeoJohnson"
for(norming in normings) {
for(trans.y in 1:1) {
df.toprocess=data.source
y.untransformed<-df.toprocess[,1]
if(norming=="centernscale"){
preProcValues= preProcess(df.toprocess[,trans.y:length(df.toprocess[1,])],method = c("center", "scale"))
df.toprocess[,trans.y:length(df.toprocess[1,])]<- predict(preProcValues, df.toprocess[,trans.y:length(df.toprocess[1,])])}
if(norming=="range01"){
preProcValues= preProcess(df.toprocess[,trans.y:length(df.toprocess[1,])],method = c("range"))
df.toprocess[,trans.y:length(df.toprocess[1,])]<- predict(preProcValues, df.toprocess[,trans.y:length(df.toprocess[1,])])}
if(norming=="expoTrans"){
preProcValues= preProcess(df.toprocess[,trans.y:length(df.toprocess[1,])],method = c("expoTrans"))
df.toprocess[,trans.y:length(df.toprocess[1,])]<- predict(preProcValues, df.toprocess[,trans.y:length(df.toprocess[1,])])}
if(norming=="YeoJohnson"){
preProcValues= preProcess(df.toprocess[,trans.y:length(df.toprocess[1,])],method = c("YeoJohnson"))#"center", "scale",
df.toprocess[,trans.y:length(df.toprocess[1,])]<- predict(preProcValues, df.toprocess[,trans.y:length(df.toprocess[1,])])}
if((norming=="asis")&&(trans.y==2)){next}
################preprocess###########
df.toprocess=data.frame(df.toprocess[dependant.selection,])
y.untransformed=y.untransformed[dependant.selection]
if(norming=="quantile"){
for(Clol in trans.y:length(data.source[1,])){
df.toprocess[,Clol]<- (rank(df.toprocess[,Clol],na.last = "keep",ties.method = "average")-1) }
preProcValues= preProcess(df.toprocess[,trans.y:length(df.toprocess[1,])],method = c("range"))
df.toprocess[,trans.y:length(df.toprocess[1,])]<- predict(preProcValues, df.toprocess[,trans.y:length(df.toprocess[1,])])}
loess.model<-loess(y.untransformed~ df.toprocess[,1],span = 0.21, degree = 1)
#df.toprocess = data.frame(df.toprocess,)
#nzv <- nearZeroVar(df.toprocess[,1:10])#, saveMetrics= TRUE
#nzv[nzv$nzv,][1:10,]
#df.toprocess = (df.toprocess[, -nzv])
tuneLength=32
tuneLength2=20
set.seed(seed.var)
inTrain <- createDataPartition(y = df.toprocess[,1],
p = .75,
list = FALSE)
training <- df.toprocess[ inTrain,]
testing <- df.toprocess[-inTrain,]
######for all models########
for(allmodel in allmodels){#just before all models define d.f and reduce it
write.table(allmodel,file = "last algorithm tried.csv", quote = F, row.names = F,col.names = F)
bad.models=c("neuralnet","partDSA","blackboost","bstSm","bstTree","penalized","brnn","gamLoess","ANFIS","FIR.DM","FS.HGD","nodeHarvest","mlpWeightDecayML","monmlp","mlp","mlpWeightDecay","mlpSGD","rbf","rbfDDA","rfRules","GFS.FR.MOGUL","mlpML","HYFIS","GFS.THRIFT" ,"GFS.LT.RS")
#too slow neuralnet
if(allmodel %in% bad.models) {next()} #gamLoess crashes. the capitals are slow and terrible
library(caret) #mlp...s creat some bizzare problem that breaks caret::train ##nodeHarvest is SLOW ##"rbf"crash R "rbfDDA" crash train and really bad #rfRules is REALLY slow.##"pythonKnnReg",pythonKnnReg can not install
#penalized slow hen fails
seed.var=seed.var+1
list.of.packages <-getModelInfo(allmodel)[[1]]$library
new.packages <- list.of.packages[!(list.of.packages %in% installed.packages()[,"Package"])]
if(length(new.packages)) install.packages(new.packages, dep = TRUE)
if(length(list.of.packages[!(list.of.packages %in% installed.packages()[,"Package"])])){
write.table(paste("Fail","Fail","Fail","Fail","PackageFail",date(),allmodel,column.to.predict,trans.y,datasource,missingdata,withextra,norming,round(proc.time()[3]-when[3]), sep = ","),
file = "gen test out.csv", append =TRUE, quote = F, sep = ",",
eol = "\n", na = "NA", dec = ".", row.names = F,
col.names = F, qmethod = "double")
next()}
when<-proc.time()
if(length(df.previous.calcs[,1])>0){
if(check.redundant(df=df.previous.calcs,norming=norming,trans.y=trans.y,withextra=withextra,missingdata=missingdata,datasource=datasource ,column.to.predict=column.to.predict,allmodel=allmodel)){next}}
not.failed=0
set.seed(seed.var)
try({trainedmodel <- train(x=data.frame(training[,2:length(training[1,])]),
y = df.toprocess[inTrain,1],
method = allmodel,
trControl = adaptControl,
tuneLength = tuneLength)
#SVMS crash cause mean needs na.rm to = T #
predicted.outcomes<-predict(trainedmodel, newdata=(testing))
p <- data.frame(predicted.outcomes,testing)
#Rsqd =(1-sum((p[,2]-p[,1])^2, na.rm = T)/sum((p[,2]-mean(p[,2]))^2, na.rm = T))
Rsqd=1-RMSE(p[,1],p[,2])/RMSE(p[,2],mean(p[,2], na.rm = T))
#mean.improvement=1-mean(abs(p[,2]-p[,1]), na.rm = T)/mean(abs(p[,2]-median(p[,2])), na.rm = T)
mean.improvement=1-MAE(p[,1],p[,2])/MAE(p[,2],mean(p[,2], na.rm = T))
p<- data.frame(predict(loess.model,predicted.outcomes),y.untransformed[-inTrain])
#RMSE=(sqrt(mean((p[,1]-p[,2])^2, na.rm = T)))
RMSE=RMSE(p[,1],p[,2])
#RMSE.mean=(sqrt(mean((p[,2]-mean(p[,2]))^2, na.rm = T)))
RMSE.mean=RMSE(p[,2],mean(p[,2], na.rm = T))
#MMAAEE=mean(abs(p[,2]-p[,1]), na.rm = T)
MMAAEE=MAE(p[,1],p[,2])
wut=print(trainedmodel,selectCol=TRUE)
overRMSE=as.numeric(wut[wut[,length(wut[1,])]=="*",length(wut[1,])-3])
replace.overRMSE=1
try({if(is.numeric(overRMSE)){replace.overRMSE=0}})
if(replace.overRMSE==1){overRMSE=-1}
if(length(overRMSE)<1){overRMSE=-1}
#print(c(Rsqd,RMSE,overRMSE,date(),allmodel,column.to.predict,datasource,missingdata,withextra,norming,adaptControl$search,seed.const,adaptControl$method,tuneLength,adaptControl$number,adaptControl$repeats,adaptControl$adaptive$min,trainedmodel$bestTune))
write.table(c(round(mean.improvement,digits = 3),round(Rsqd,digits = 3),round(overRMSE,digits = 3),round(RMSE,digits = 3),round(MMAAEE,digits = 3),date(),allmodel,column.to.predict,trans.y,datasource,missingdata,withextra,norming,RMSE.mean,adaptControl$search,seed.var,round(proc.time()[3]-when[3]),adaptControl$method,tuneLength,adaptControl$number,adaptControl$repeats,adaptControl$adaptive$min,trainedmodel$bestTune),
file = "gen test out.csv", append =TRUE, quote = F, sep = ",",
eol = "\n", na = "NA", dec = ".", row.names = F,
col.names = F, qmethod = "double")
print(date())
not.failed=1
})
if(not.failed==0) {
try({trainedmodel <- train(x=data.frame(training[,2:length(training[1,])]),
y = df.toprocess[inTrain,1],
method = allmodel,
trControl = simpleControl,
tuneLength = tuneLength2)
predicted.outcomes<-predict(trainedmodel, newdata=(testing))
p <- data.frame(predicted.outcomes,testing)
#Rsqd =(1-sum((p[,2]-p[,1])^2, na.rm = T)/sum((p[,2]-mean(p[,2]))^2, na.rm = T))
Rsqd=1-RMSE(p[,1],p[,2])/RMSE(p[,2],mean(p[,2], na.rm = T))
#mean.improvement=1-mean(abs(p[,2]-p[,1]), na.rm = T)/mean(abs(p[,2]-median(p[,2])), na.rm = T)
mean.improvement=1-MAE(p[,1],p[,2])/MAE(p[,2],mean(p[,2], na.rm = T))
p<- data.frame(predict(loess.model,predicted.outcomes),y.untransformed[-inTrain])
#RMSE=(sqrt(mean((p[,1]-p[,2])^2, na.rm = T)))
RMSE=RMSE(p[,1],p[,2])
#RMSE.mean=(sqrt(mean((p[,2]-mean(p[,2]))^2, na.rm = T)))
RMSE.mean=RMSE(p[,2],mean(p[,2], na.rm = T))
#MMAAEE=mean(abs(p[,2]-p[,1]), na.rm = T)
MMAAEE=MAE(p[,1],p[,2])
print(confusionMatrix(p[,1],p[,2]))
wut=print(trainedmodel,selectCol=TRUE)
overRMSE=as.numeric(wut[wut[,length(wut[1,])]=="*",length(wut[1,])-3])
replace.overRMSE=1
try({if(is.numeric(overRMSE)){replace.overRMSE=0}})
if(replace.overRMSE==1){overRMSE=-1}
if(length(overRMSE)<1){overRMSE=-1}
#print(c(Rsqd,RMSE,overRMSE,date(),allmodel,column.to.predict,datasource,missingdata,withextra,norming,adaptControl$search,seed.const,adaptControl$method,tuneLength,adaptControl$number,adaptControl$repeats,adaptControl$adaptive$min,trainedmodel$bestTune))
write.table(c(round(mean.improvement,digits = 3),round(Rsqd,digits = 3),round(overRMSE,digits = 3),round(RMSE,digits = 3),round(MMAAEE,digits = 3),date(),allmodel,column.to.predict,trans.y,datasource,missingdata,withextra,norming,RMSE.mean,simpleControl$search,seed.var,round(proc.time()[3]-when[3]),simpleControl$method,tuneLength2,simpleControl$number,"no rep","no min",trainedmodel$bestTune),
file = "gen test out.csv", append =TRUE, quote = F, sep = ",",
eol = "\n", na = "NA", dec = ".", row.names = F,
col.names = F, qmethod = "double")
print(date())
not.failed=1
})
}
if(not.failed==0) {
print(c("failed","failed",date(),datasource,missingdata,withextra,norming,allmodel))
write.table(paste("Fail","Fail","Fail","Fail","Fail",date(),allmodel,column.to.predict,trans.y,datasource,missingdata,withextra,norming,round(proc.time()[3]-when[3]), sep = ","),
file = "gen test out.csv", append =TRUE, quote = F, sep = ",",
eol = "\n", na = "NA", dec = ".", row.names = F,
col.names = F, qmethod = "double")
write.table(paste("Fail",date(),allmodel, sep = ", "),
file = "backup.csv", append =TRUE, quote = F, sep = ",",
eol = "\n", na = "NA", dec = ".", row.names = F,
col.names = F, qmethod = "double")
}
if(not.failed==1) {
write.table(paste("Succ",date(),allmodel, sep = ", "),
file = "backup.csv", append =TRUE, quote = F, sep = ",",
eol = "\n", na = "NA", dec = ".", row.names = F,
col.names = F, qmethod = "double")}
}}}}}}
#stopCluster(cl)
## reset message sink and close the file connection
sink(type="message")
close(zz)
### Display the log file
readLines("all.Rout")