#read rhdf file library(rhdf5) train <- h5read("../input/train.h5","train") columns <- train$axis0 train <- data.frame(as.double(train$block0_values[1,]), as.double(train$block0_values[2,]), t(train$block1_values)) colnames(train) <- columns H5close() #taken test 2% of test data and remove column 1 and 2 because ID and timestamp testsize<-nrow(train)*0.02 testframe<-train[c(1:abs(testsize)),c(-1,-2)] # cleaning process replace empty values with zero testframe[is.na(testframe)]<-0 #testframe some of box plot to observe skewness boxplot(testframe[1:500,]) # apply normalization min-max scaling test.mm <- apply(testframe[,-109], MARGIN = 2, FUN = function(x) (x - min(x))/diff(range(x))) boxplot(test.mm[,c(1:10)]) # z-score test.mm <- apply(testframe[,-109], MARGIN = 2, FUN = function(x) (x - mean(x))/sd(x)) test.mm <- apply(testframe, MARGIN = 2, FUN = function(x) (x - mean(x))/(2*sd(x))) boxplot (test.mm[,c(1:10)], main = "Z-score, 1 sd") boxplot (test.mm[,c(1:10)], main = "Z-score, 2 sd") #-----Section 09------------------------------------------- # soft max scaling - iterative library(DMwR) #lambda scaling test.lm <- apply(testframe[-109], MARGIN = 2, FUN = function(x) (SoftMax(x,lambda = 4, mean(x), sd(x)))) boxplot (test.lm , main = "Soft Max, lambda = 4") test.lm <- apply(testframe[-109], MARGIN = 2, FUN = function(x) (SoftMax(x,lambda = 6, mean(x), sd(x)))) boxplot (test.lm , main = "Soft Max, lambda = 10") test.lm <- apply(testframe[,-109], MARGIN = 2, FUN = function(x) (SoftMax(x,lambda = 20, mean(x), sd(x)))) boxplot (test.lm , main = "Soft Max, lambda = 20") #min-max modelframe<-data.frame(test.mm) #lambda modelframe<-data.frame(test.lm) #all na clean modelframe[is.na(modelframe)]<-0 summary (modelframe) #regression predictors modelframe$y<-testframe$y n<-names(modelframe) f<- as.formula(paste("y ~",paste(n[!n %in% "y"],collapse = " + "))) reg<-lm(f,data = modelframe) summary(reg) #all signifcant min / max modelrefined<-data.frame(modelframe$fundamental_1,modelframe$fundamental_7,modelframe$fundamental_18,modelframe$fundamental_20,modelframe$fundamental_21,modelframe$fundamental_24,modelframe$fundamental_25,modelframe$fundamental_34,modelframe$fundamental_36,modelframe$fundamental_41,modelframe$fundamental_43,modelframe$fundamental_52,modelframe$fundamental_54,modelframe$technical_0,modelframe$technical_2,modelframe$technical_6,modelframe$technical_11,modelframe$technical_13,modelframe$technical_17,modelframe$technical_18,modelframe$technical_19,modelframe$technical_20,modelframe$technical_21,modelframe$technical_22,modelframe$technical_30,modelframe$technical_36,modelframe$technical_40,modelframe$technical_43) boxplot(modelrefined) #all signifcant z score modelrefined<-data.frame(modelframe$fundamental_5,modelframe$fundamental_7,modelframe$fundamental_10,modelframe$fundamental_25,modelframe$fundamental_26,modelframe$fundamental_27,modelframe$fundamental_31,modelframe$fundamental_48,modelframe$fundamental_51,modelframe$fundamental_53,modelframe$fundamental_54,modelframe$technical_2,modelframe$technical_6,modelframe$technical_13,modelframe$technical_17,modelframe$technical_18,modelframe$technical_19,modelframe$technical_20,modelframe$technical_21,modelframe$technical_30,modelframe$technical_37) #all signifcant lambda 4 modelrefined<-data.frame(modelframe$fundamental_5,modelframe$fundamental_7,modelframe$fundamental_10,modelframe$fundamental_25,modelframe$fundamental_26,modelframe$fundamental_26,modelframe$fundamental_27,modelframe$fundamental_31,modelframe$fundamental_41,modelframe$fundamental_48,modelframe$fundamental_51,modelframe$fundamental_53,modelframe$fundamental_54,modelframe$technical_2,modelframe$technical_13,modelframe$technical_17,modelframe$technical_19,modelframe$technical_20,modelframe$technical_21,modelframe$technical_30,modelframe$technical_36,modelframe$technical_41) #all signifcant lambda 20 modelrefined<-data.frame(modelframe$fundamental_10,modelframe$fundamental_25,modelframe$fundamental_41,modelframe$fundamental_43,modelframe$fundamental_48,modelframe$fundamental_51,modelframe$fundamental_53,modelframe$fundamental_54,modelframe$fundamental_60,modelframe$technical_0,modelframe$technical_2,modelframe$technical_6,modelframe$technical_11,modelframe$technical_13,modelframe$technical_17,modelframe$technical_20,modelframe$technical_21,modelframe$technical_22,modelframe$technical_30,modelframe$technical_36,modelframe$technical_40) # give to other predicted model to check significance library(neuralnet) # simple ANN with only a single hidden neuron n<-names(modelrefined) modelrefined1<-round(modelrefined,digits=2) modelrefined1 d<-modelframe$y f<- as.formula(paste("d~",paste(n[!n %in% "y"],collapse = " + "))) #scaled data nn <- neuralnet(f,data=modelrefined1,stepmax = 1e+05,threshold = 0.01, hidden=2,learningrate=2,algorithm = "rprop+") #modelframe$y~modelframe$fundamental_10+modelframe$fundamental_18+modelframe$fundamental_25+modelframe$fundamental_33+modelframe$fundamental_43+modelframe$fundamental_51+modelframe$fundamental_52+modelframe$fundamental_62+modelframe$technical_2+modelframe$technical_13+modelframe$technical_19+modelframe$technical_21+modelframe$technical_22+modelframe$technical_36+modelframe$technical_40, data=modelrefined1,stepmax = 1e+05,threshold = 0.01, hidden=2,learningrate=2,algorithm = "sag" ) # visualize the network topology plot(nn) #this repeatedly improve the result when hidden nodes increased res<-compute(nn,modelrefined) # obtain predicted predict1 <- res$net.result # examine the relationship between predicted and actual values plot(predict1, modelframe$y) cor(predict1, modelframe$y) #- hidden node at 2 it gives correlation of 0.11 but if tested with more #-- nodes the correlation at 5 hidden layers is 0.16, it eventually #-- deduced that it require a continues learning process into deep #-- layers with adusted weight precautionary ,the conclusion gives reader a #-- standing direction of reinforcement learning with more layers. #-- Min - Max at hidden layers 2 give 0.11 correlation #-- z at hidden layers 2 give 0.13 result #-- 0.11 is at lambda 4 #-- 0.093 is at lambda 20 rm(list=ls())